High school | For students

Artificial Intelligence E t h ics

Introduction

Artificial intelligence (AI) ethics is evolving in line with the development of AI. AI can have a positive or a negative influence on society, depending on how we develop and use it. In the near future, we need to develop an ethical perspective based on an understanding of the technical principles of AI. This is because social problems we will encounter in the age of AI will be different from the ethical problems we are currently facing in human society.

Thus, demand for AI ethics education is gradually increasing. This change stems from the increased use of AI technology in our lives and the direct impact of AI on our social and economic spheres. Developing an AI ethic is the process of forming a social consensus to use AI in beneficial ways through efforts to maximize its positive effects and reduce any negative influence.

AI is being developed and used for the purpose of improving the quality and convenience of our lives. As with all other technologies, although AI is developed for the benefit of people, adverse effects may become apparent in the course of developing and using AI. However, AI has infinite possibilities in that it can judge and act in consideration of surrounding situations beyond the level of human intelligence, in contrast to existing automated systems that only repeat predetermined routines. Because AI is generating such high expectations and fears at the same time, AI ethics is more important than ever.

In the beginning, many thought that ethics should be programmed into AI itself. However, as people have come to understand that humans are ultimately responsible for creating and using AI, the importance of understanding AI ethics is now at the forefront of many people’s minds. For now, students are mainly interacting with AI as users, but they may grow to become AI developers or managers in the future. Therefore, it is crucial to educate students going forward to help them think about AI ethics and put them into practice. These students will be the ones to participate in the overall process of developing and using AI in the future.

The development of AI technology has been projected to usher in the era of the ‘Fourth Industrial Revolution.’ Such technical advancement also sparks the fear that humans may be replaced by machines or the anxiety that AI could harm humans by deviating from its intended purpose and being misused, similar to the fears of coachmen who refused the change when the automobile was first invented.

It is not reasonable to go against the tide of the ‘Fourth Industrial Revolution’ that has already arrived due to our vague fears and anxiety about AI. Therefore, we need to start our discussions about ‘human-centered AI’ and ‘trustworthy AI’ now.

Based on the 10 key requirements of 「Artificial Intelligence Ethical Standards」, this book is designed to facilitate the analysis of AI and a range of related ethical issues from various perspectives in the educational field. In addition, in contrast to the game-based book for elementary schools and experience-based book for those in middle school, this book provides an opportunity for high school students to explore AI ethics in depth.

School education in the age of AI should address the issues of how to coexist with AI and how to ensure that human values are taken into account. Knowledge of AI’s basic working principles, fundamentals, and techniques is already covered in existing curricula. However, education about the preparation required for AI and the values its users should pursue is still insufficient.

Throughout this book, the writers intended to help students expand their intellectual horizons by balancing the positive and negative aspects (limitations, excessive expectations of AI, biased AI systems, fake news, influences on privacy and employment, etc.) that AI can bring to our society.

This book also aims to supplement numerous textbooks explaining AI basics. Thus, this book does not start with basic questions, including “What is AI?” “What is a neural network?” “How are AIS developed and trained?” “What are some ways AI is used today?” and “What are some future directions for AI?” In this regard, the intended users of this book are students who already have a basic understanding of how AI works. Although programming skills are not prerequisites for students to study this book, some chapters contain simple programming exercises. However, the authors make these activities easy to follow so that students with minimal knowledge of AI can learn the lessons.

We hope that this book will provide an opportunity for students to grow into global citizens who actively respond to the world’s problems and create a sustainable society. We also hope that students will develop an understanding of AI technology and recognize the existence of multiple standards for ethics by looking at a range of cases related to AI. We will look at cases including issues of responsibility for self-driving cars, determining who should be held accountable in case of any damage caused by AI malfunctions, biased algorithms, and others. Finally, we hope that students will develop a competency to solve basic and advanced research problems in AI.

Korea Information Society Development Institute (KISDI)

Ⅲ

Learning Objectives

Safeguarding Human Rights

Protecting Privacy

Respecting Diversity

Doing No Harm

Working for the Common Good

Strengthening Solidarity

Managing Data Properly

Taking Responsibility

Ensuring Safety

hisLessonInT

Maintaining Transparency

Structure and Features of This Book

Think about the Problem

Respecting Diversity

AI systems might follow and reproduce the stereotypes and prejudices inherent in our current society, resulting in issues related to bias, appropriateness, and fairness. What efforts should be made to prevent these issues?

You will be able to understand the concept and importance of respecting diversity in the context of using AI. You will be able to explain the need to resolve discrimination and bias in the development and use of AI systems in relation to the principle of protecting privacy. You will be able to suggest ways to respect diversity in the deployment of AI technologies and services.

Discover Problems in Our Daily Lives

What issues can Somin bring up in the following situation?

I am thinking about going to the movies

this Saturday. Can you recommend

guys going to What are you I am not any good movies? Well … These days, you can That is a good idea.

the recommended You can check out

sure.

optimized and customized get recommendations for

do tomorrow? It is Saturday!

movies.

content from algorithms that predict what you like. How

about that?

I haven’t the algorithms thought about algorithm-based

that way. What do you think, of algorithm- The problem Somin?

Yeah, but the algorithms only similar movies. It gets recommend all these

based personalized content …

boring.

tailored content I agree. Those

by AI algorithms recommended

rather makes me bored.

Write down your opinion.

You will recognize the need for diversity in every area where AI technology is used and explore measures to ensure the protection of privacy.

Think about the Problem

Start the lesson

Respecting Diversity

is a problem with Well. Maybe that

personalized content.

STEP 1 Get a glimpse of the learning content Look at the question and pictures of ‘Discover problems with pictures and learning objectives in our daily lives’ related to the 10 requirements of the Artificial Intelligence Ethical Standards, write down your associated with the lesson. thoughts on them, and predict what you will learn.

ThisLessonnI

Ⅳ

Explore Key Concepts

Examine the concept of no violation, one of the 10 key requirements.

What is ‘no violation’ in the context of AI?

No violation means that we should pay more heed to the prevention of any infringement on the safety of our society and the rights of individuals in developing and using AI, and to predict and prevent infringements in advance. AI system is a technology developed for the benefits of humans. Therefore, AI systems should be designed to prevent the infringement of basic human right is infringed through prior investigation of various types of potential damage, including physical and mental harm, in the development phase. Even if AI systems are designed with possible infringement of human rights considered, they may cause damage to users if they are used beyond their intended purpose or if direct or indirect risks occur due to unintended or unexpected factors. Hence, we should seek solutions to address damages from AI and minimize the potential risk factors of AI system by looking at various cases of damage caused by AI systems.

Use only for intended

purposes

basic human Violation of

Examine any potential damage to users

rights

What else should be taken into account?

Types of violations from AI 2

A consulting company in the US has categorized five types of violations that may take place due to errors or misuse of AI systems.

Failure of management

Misuse of data

This is a type of infringement caused by misuse of 2 1 data for AI, violation or misinterpretation of security

This is a type of violations that occurs in relation to various negligence and improper management that can be caused by the AI system manager in using AI systems.

regulations, and occurs when legal provisions are violated in collecting and using data.

I am sure AI finished all the work

Stop it! You should observe the regulations.

Explore Key Concepts

STEP 2 STEP 3 Delve into the concepts of the 10 Take a look at ethical cases in various key requirements of the artificial fields where AI is deployed, and intelligence ethical standards from discuss the ideas for cooperation social and humanistic perspectives, based on what you learned in including the impact of AI on society and its interactions with humans.

Doing No Harm

Learning Objectives

학

You will be able to look at various violation cases related to AI systems and find solutions for them. You will be able to understand that developers, users, and policy makers in the event of any violation due to the use of AI systems, need to make shared efforts to compensate for damages and prevent any recurrence.

Developer

maker Policy

User

Safeguarding Human Rights

Protecting Privacy

Respecting Diversity

Doing No Harm

Working for the Common Good

Strengthening Solidarity

Managing Data Properly

Taking Responsibility

Ensuring Safety

InThisLeso

n

Maintaining Transparency

You will learn that AI should not be used to harm humans, and discover ways to deal with the risks and negative consequences that can be triggered by AI.

Cases

Check key concepts with

STEP 2

Basic Problem Solving Activity

기초 활동 문제 해결하기

Determine who should take the responsibility for AI systems

Activity 1

Taking Responsibility

Give your opinion on who should be held accountable for AI systems in the following situation. Also, if the court rules that the driver should take responsibility, predict the future impact of this judgment on human society.

Fatal accident due to autopilot

A driver who was involved in a fatal accident while driving a car on autopilot, a self-driving assisting function of a car from Company T, has been charged with murder. This is the first time in Country A that the crime of murder has been applied to a normal driver who used the self-driving function. According to foreign press, the prosecutors of Country A indicted Mr. K on charges of accidental murder for crashing into an oncoming car while driving on autopilot, resulting in the death of two people. Mr. K had a criminal record of speeding at an intersection two years ago in December 2019 by driving on autopilot and ignoring traffic lights. Autopilot is a function that assists the driver in terms of accelerating, braking, and steering the car. Currently, the number of vehicles equipped with this function in Country A is about 765,000. Autopilot is a function that assists the driver, due to the misperception of some drivers who misunderstand it as a 'fully autonomous driving function', there has been a string of related accidents. Currently, 26 autopilot-related accidents that took place after 2016 are being investigated by the Road Traffic Safety Bureau of Country A. These accidents were reported to have taken the lives of at least 11 people. An official from the Road Traffic Safety Bureau commented, “There are no cars available in the market yet that are equipped with full self- driving functions. All cars, with or without self-driving functions, must always be controlled by a human driver.”

Driver

Autopilot

<Source> The Chosun Ilbo, Sep. 01, 2019

Opinion on who should take the responsibility

Prediction of the influences on human society

Problem Solving Activity

Wrap up the Lesson

Self Checklist

Strengthening Solidarity

1. Mark with O for correct statement and X for incorrect statements.

1 According to the definition from dictionaries, solidarity refers to 'the characteristics of being connected to each other as one unit'. However, when etymologically considered, the word came from the Latin word ‘solidare’, which means ‘to firmly unite’. Accordingly, solidarity implies the meaning of firmly uniting, and in this case, bonding is an essential characteristic of each human being, and indicates the mutual combination of personality and sociality. ( ) 2 Cooperations in the international community to maintain relational solidarity among various groups and build an AI system that is fully considerate of future generations are not required.

( )

2. Take initial letter quizzes and review key terms.

1 Hans Jonas stated “An actor should assume the responsibility for his/her act. Even if the cause was not evil, and even if the result was not expected or intended, the harm caused should be compensated for” to emphasize [E ] of [R ].

2 At the 70th session of the UN General Assembly held in 2015, the international community unanimously adopted

17 [S ] [D ] [G ] to achieve sustainable development, including no poverty, zero hunger, and reduced inequality.

3. Briefly describe why international agreement and cooperation are required in the development and use of AI.

책임성

Advanced Problem Solving Activity

Analyze the case related to face recognition technology

Activity 1

Doing no harm

Analyze the case of an incidents caused by errors in the face recognition system, and think about the rights violated and the proper remedies therefor.

In January 2020, Mr. A, a black man living in Detroit, was spotted as a suspect of a watch theft from a luxury store. In this case, he was arrested by the police in the front yard of the house with his wife and daughters looking the scene, but it later turned out that he was innocent, but he had been arrested due to an error in the Detroit Police Department's face recognition system.

Think about the case from the respective perspectives of Mr. A and the police, the parties concerned, and write your opinion.

Case Overview

innocent. No, I am You are a In March 2019, the Detroit Police Department put criminal. You are under a fuzzy photo of a black man in a red hat from arrest. CCTV into a face recognition system developed by Company D to contrast it with photos from driver's licenses stored in national photo repositories across the United States. As a result, it was discovered by the system that the photo of the suspect and the photo of Mr. A's driver's license matched. Afterwards, the investigators added Mr. A to the list of suspects, and the watch store's security officer, who was not present at the time of the theft, identified Mr. A as the culprit of the theft. Detectives issued a warrant that required the signature of a high-ranking executive of the police department and arrested Mr. A. After checking the photo of a suspect presented by a detective of the Detroit Police Department during interrogation, Mr. A said, “This is not me. I hope you don't think all black people look alike.” However, the detective replied “The computer says it’s you.” Consequently, Mr. A had to spend 30 hours in a detention center. I hope you don't think all black people

look alike.

If I were Mr. A?

If I were a police officer?

Analyze the points in dispute of the case and the positions of both sides, and think about the violation of rights and applicable remedies under Korean legal provisions

Main Issues of the Case The Detroit Police Department relied on the biased and flawed face recognition technology to arrest Mr. A. Mr. A's basic human rights guaranteed equally to everyone were violated due to his race or skin color, and flaws in the use of AI technology were revealed.

STEP 4 STEP 5 Develop ethical thinking skills and problem-solving skills through a range of activities covering basic and advanced levels, and apply these skills in real life.

4. Review the takeaways of the lesson you have learned with a mind map.

Strengthening Solidarity

Solidarity

Criteria

You can explain the concept and meaning of solidarity with respect to the use of AI. You can predict the impact that the use of AI will have on our future generations and propose adequate solutions.

You can suggest the desirable role and scope of AI technology to make sure it contributes to the integration and development of society.

Harmonious co-existence of

Importance of the Right to Self- determination of Personal Information

AI and Us

J.S. Mill, a British liberal thinker, stated in his work On Liberty, “Over himself, over his own body and mind, the individual is sovereign.” This sentence can be reinterpreted to mean “in the era of AI, everyone has the right to control his/her personal information”. The right to control personal information is also called the “right to self-determination of personal information.” The ‘right to self-determination of personal information’, which refers to the information subject’s right to make a decision on his/her own concerning the disclosure and use of his/her personal information, can be justified on the grounds of basic human rights stipulated in the Constitution. That is, the right to self-determination of personal information can be guaranteed on the basis of the right to pursue happiness with human dignity and value and the right not to be infringed on the secrecy and freedom of privacy as prescribed in the Constitution. This right allows everyone to decide for himself/ herself when, to whom, and to what extent information about him/her is made available and used.

Protecting Privacy

I will decide to allow the use of this part of my personal information only over one year

I need your personal information.

Government, private companies and AI systems

Personal information

Developer

I will control and make decisions The discussion on the ‘right to self-determination of personal information’ has grown in about my personal information on my importance with the advent of the information age. Together with values such as the own. 'right to know' and the 'freedom of expression', this right is triggering various ethical debates. As science and technology become more advanced, it is highly likely that the current personal information consent system will not be applied in practice. This is because in the advanced form of the Internet of Things (IOT) system, in which the collection and use of information is carried out without human intervention, it can be very difficult to obtain prior consent from each individual for the collection and use of personal information. Therefore, individuals, companies, and countries are required to draw a social consensus on the ways to efficiently develop and utilize AI technology while protecting privacy.

AI and Us

Harmonious co-existence of STEP 6 Review the learning contents through a range of activities such as true and Present interesting stories and real-life false quizzes, initial letter quizzes, drawing line games, and mind maps, and cases related to AI to stimulate and satisfy self-check the learning processes.

curiosity.

Table of Contents

Intro.

Ⅰ

Safeguarding Human Rights

Check Key Concepts with Cases 18

Basic Problem Solving Activity 20

Advanced Problem Solving Activity 24

Wrap up the Lesson

STEP 6

Ⅲ

Respecting Diversity

STEP 1

STEP 2

Check Key Concepts with Cases 54

STEP 3

Basic Problem Solving Activity 55

STEP 4 STEP 5 Harmonious co-existence of

Necessity of Education on AI Ethics

Ⅱ

Protecting Privacy

AI and Us

STEP 6

Ⅳ

STEP 1

STEP 2 STEP 3 STEP 4

Advanced Problem Solving Activity 58

STEP 5

AI and Us

AI and Us

35 Check Key Concepts with Cases 36 Basic Problem Solving Activity 38 Advanced Problem Solving Activity 44

66 Check Key Concepts with Cases 69 Basic Problem Solving Activity 70 Advanced Problem Solving Activity 72

Ⅴ

Working for the Common Good

Think about the Problem

STEP 1

Explore Key Concepts

STEP 2 STEP 3 STEP 4 STEP 5

Wrap up the Lesson

STEP 6

Ⅶ

Managing Data Properly

Wrap up the Lesson

STEP 6

Ⅸ

Ensuring Safety

STEP 1

STEP 2 STEP 3 STEP 4 STEP 5

Ⅵ

STEP 1

STEP 2

Check Key Concepts with Cases 84

STEP 3

Basic Problem Solving Activity 86

STEP 4

Advanced Problem Solving Activity 92

STEP 5

STEP 6

Ⅷ

Taking Responsibility

Check Key Concepts with Cases 125

Basic Problem Solving Activity 126 Advanced Problem Solving Activity 136

STEP 6

Ⅹ

STEP 1

STEP 2

Check Key Concepts with Cases 174

STEP 3

Basic Problem Solving Activity 176

STEP 4

Advanced Problem Solving Activity 180

STEP 5

• Index

Strengthening Solidarity

Think about the Problem

Explore Key Concepts

102 Check Key Concepts with Cases 104 Basic Problem Solving Activity 106 Advanced Problem Solving Activity 110 AI and Us 116

Harmonious co-existence of

Wrap up the Lesson

152 Check Key Concepts with Cases 154 AI and Us 155

Harmonious co-existence of

Basic Problem Solving Activity 156 Advanced Problem Solving Activity 160 Wrap up the Lesson 168

Maintaining Transparency

190 Check Key Concepts with Cases 193 Basic Problem Solving Activity 194 AI and Us 197

Harmonious co-existence of

Advanced Problem Solving Activity 198

Wrap up the Lesson

204 • Photo Sources, References and Citations 206

Necessity of Education on AI Ethics

Why is AI Ethics Important?

There is a high probability that AI is already embedded in the mobile phones, computers, cars, etc. that we use today. Although self-driving cars are not widely used at the moment, it is expected that in the years to come, AI-powered cars that detect the movement of objects by observing the flow of traffic in a city will begin to be used. In addition, AI will be widely used for chatbot control or financial asset management and investment through apps, and AI may be more widely used for conducting job interviews. AI will be used not only in the family domain, but also in the process of police investigation of criminals and finding missing people.

AI can have a positive or a negative influence, depending on how humans develop and use it. As the use of AI becomes prevalent in recent years, the controversy over the impact of AI on society is also intensified.

Page 35 Privacy Protection

Timeline for the Development of AI Ethics

DeepMind Ethics & Society Principles

(DeepMind) Asilomar AI Principles

AI Principles (Microsoft, Google, IBM)

(Future of Life Institute)

Solidarity on Page 103

Recommendation of the Council on Artificial Intelligence (OECD) Ethics Guidelines for Trustworthy Artificial Assessment List for Trustworthy Artificial Intelligence ) Intelligence ) (EU (EU AI Fairness Checklist Everyday Ethics for AI (Microsoft) (IBM) Understanding AI Ethics and Safety Rome Call for AI Ethics

(The Alan Turing Institute) (The Vatican)

The Common Good on Page 82

Civil Society Private Companies International Organizations Korean Government Religious Organizations or Public Institutions Foreign Governments Non-profit Organizations

Recommendation on the Ethics of Artificial AI Bill of Rights

(The White House)

Intelligence (UNESCO)

Facing this situation, the international community is becoming more vocal about “Artificial Intelligence Ethics.” As part of these efforts, the European Commission (EC) published the <Ethical Guidelines for Trustworthy Artificial Intelligence> in 2019, and proposed the ‘Artificial Intelligence Act’ in 2021. Ethical guidelines teach businesses and citizens the important elements in using AI and provide guidelines for developers on how to develop AI that benefits humans. In addition, the Artificial Intelligence Act embodies the perception that regulation of AI systems should be performed in phases according to the degree of risk.

In 2022, UNESCO established international standards with normative effect on AI and based on active cooperation from member states, published the “Recommendations on the Ethics of Artificial Intelligence” to help countries use AI for the benefit of humans. Korea is also making steady efforts to establish ethical principles for AI. Since the initiation of the formulation of AI ethical principles for the private sector in 2018, the Republic of Korea announced the human-centered <Artificial Intelligence (AI) Ethical Standards> in 2020 at the governmental level, and <National Strategy for Artificial Intelligence> in the following year.

Protection of Privacy on Page 33

Korea

Ethical Guidelines for Intelligent Information Society Robotic Ethics

(Information and Culture Forum) AI Algorithm Ethics Charter

Artificial Intelligence Ethics Charter Ethical Guidelines on

(Kakao)

Respecting Diversity on Page 54

National Strategy for Artificial Intelligence

media recommendation services

(Korean government)

Principles for the User-oriented Intelligent

Information Society (Korea

Communications Commission)

Human-centered Artificial Intelligence

Basic Principles of

Ethical Standards (Korean government)

(Ministry of

Trade, Industry and Energy)

Self-Driving Vehicles (Ministry of Democracy (120 civic groups in

(International Association for Artificial Intelligence and Ethics) Land, Infrastructure and Transport) Korea)

Responsibility on Page 156

Driver

Autopilot

National Strategy for Artificial Intelligence

(Korean government)

Basic principles for the protection of users of AI-based

Human Rights Guidelines for the Development and Utilization of

Artificial Intelligence (National Human (Korea Communications Rights Commission of the Republic of Korea) Commission) Artificial Intelligence Ethical Principles in NAVER AI Code of Ethics , Education (Ministry of AI Values to Pursue (Naver) Education) (SKT) AI Ethical Principles Civil Societies’ Declaration Demanding Artificial

(LG)

Intelligence Policies to Ensure Human Rights, Safety, and AI Chatbot Code

of Ethics

(ScatterLab)

Why Should We Learn AI Ethics?

Since the ethical issues that arise in the era of AI differ from the moral issues in the existing human society, we must understand both the technical principles and ethical standards of AI. Due to its technical nature, AI can learn on its own without human intervention, and in many cases humans cannot accurately predict the outcome of its judgments. In the future, as more diverse AI technologies are predicted to directly affect our daily lives, social and economic benefits, and opportunities in life, we need to pay closer attention to how AI technologies are used.

“

In line with rapidly developing science and technology, the current generation has the duty to exercise responsibility towards its descendants, a move that requires ethics. - Hans Jonas

”

As the remarks of Hans Jonas, a German philosopher who emphasized responsibility ethics, suggest, the development and use of AI technology has a great impact not only on individual lives and society, but also on the environment. Learning AI ethics is intended to maintain human dignity in the era of AI through the efforts to maximize the positive functions of AI technology and minimize the negative ones.

What Should We Learn about AI Ethics?

The book on <AI ethics> guides you to identify ethical issues by exploring AI use cases across the globe based on the understanding of the concept of AI, its principles, and how it works. In this course, students will learn AI ethics based on the 3 basic principles and 10 key requirements of the <Artificial Intelligence Ethical Standards> published by the Korean government.

Basic Principle 1

Human Dignity

As living creatures and beings with personalities, human beings have inherent dignity, unlike technology and tools developed for humans. In human history, there have been cases in which human dignity has been infringed by others or institutions, but in the upcoming AI era, new infringements may occur unintentionally in the process of developing and utilizing AI systems. Therefore, AI systems should be developed in a way that respects and protects fundamental human dignity.

Basic Principle 2

Common Good of Society

Basic Principle 3

Proper Use of Technology

What if the benefits of AI were only enjoyed by a select few? In the knowledge and information society of the future, universal access to AI must be guaranteed even for the socially underprivileged or vulnerable groups who are likely to be alienated. We must keep an eye on the risk of the weakened democracy or undermined rule of law due to AI technology, and ensure universal human welfare at the national and global levels. To this end, humans who develop and use AI should actively pursue the common good of society.

We need to preempt the impact that AI could have. Positive expectations for AI can lead to opposite results, and negative concerns might actually hold back innovations that make human life easier. The point we need to keep in mind is that AI technology should be developed and used to serve its intended purpose and intention as a necessary tool for human life, and that the process of development and use itself should be ethical as well. Therefore, those who develop and use AI technology should make concerted efforts to secure technical and institutional mechanisms to prevent misuse of AI technology or the disproportionate distribution of relevant benefits to specific groups.

In order to prevent AI technology from being misused or abused, undermining human values and ethics, we must develop the ethical competency to determine the problems of AI system and causes of and solutions to such problems.

Benefits of Education on AI Ethics

Today, as the development and use of AI technology accelerates in all areas of society, the unique roles of relevant actors such as governments, businesses, and citizens are becoming more important. Problems that damage public values or individual ethical values may occur in the process of pursuing economic profit or improving work efficiency using AI. Therefore, more weight should be given to discussions on corporate social responsibility or governmental public values in developing AI. Furthermore, a forum for free debates and discussions among citizens, governments, and businesses should be prepared. In other words, citizens or experts with interest in the development of AI technology

should be able to participate in public discussions, express their preferences, and formulate policies or ethical standards in the public interest.

of AI systems

Active discussions on the use of AI technology can promote the development of democracy. Moreover, citizens can create a “human-centered AI world” through active participation and discussion on current issues related to AI ethics.

Let’s begin our journey to systematic learning about AI as well as the importance of extensive participation and ethics, bearing in mind the importance of discussions cooperation among various stakeholders as highlighted and consensus regarding the management and regulation in the <Recommendation on the Ethics of Artificial Intelligence> by UNESCO.

Ⅰ

Safeguarding Human Rights

Learning Objectives

You will be able to understand that the violation of human rights and freedoms should be avoided in the development and use of AI. You will be able to explain that it is important to make sure that AI technologies are used in a way that respects everyone’s rights and our democratic values. You will be able to suggest technical, institutional, and ethical measures to ensure that AI is used fairly and responsibly.

Right to know Right not to be

discriminated against

Equality

Basic human rights Human

Right to be forgotten

You will understand the concept of protection of human rights, investigate cases in which fundamental human rights are violated by AI, and present plans to guarantee the freedoms and rights that all humans should enjoy at the technical, institutional, and ethical levels.

Freedom

UN

rights

Right to privacy

‌ ‌AI technology makes human life convenient and increases work efficiency, but there are cases where human rights are not fully protected in the process of using AI. What can we do to safeguard human rights?

I have to find a way to get this the article about the seizure case from 10 years ago wiped from the internet!

No way, even though my property was seized and I paid off my debts 10 years ago, you can still read about it online. What must people reading this article think about me?

‌ ‌Examine the concept of Safeguarding Human Rights, one of the 10 key requirements.

What is Safeguarding Human Rights?

Every human being has the right to be treated respectfully. Humans should be recognized for the absolute value of their existence and not be treated as a means to an end. These rights that every human being deserves are called human rights. Human rights are rights inherent in human beings from the moment they are born, and are naturally granted to human beings even before they are guaranteed by institutions such as the Constitution or laws. For this reason, human rights are also referred to specifically as natural rights*. Human rights are characterized by being possessed by every person across the globe, non- transferable and granted permanently from birth.

Legal provisions for the Safeguarding Human Rights

People come to reveal their own unique characteristics mentally and physically from birth or during their journey through life. They diverge in values in society and life, and have different skin colors, religions, and genders. Some have disabilities, and others have great intelligence. Despite all these differences, all people are guaranteed equal rights as long as they are born human. This principle is protected by the Universal Declaration of Human Rights* and the Constitutions of each country.

Article 1 of the Universal Declaration of Human Rights

All human beings are born free and equal in dignity and rights. They are endowed with reason and conscience and should act towards one another in a spirit of brotherhood.

The Universal Declaration of Human Rights sets forth that every human being is born free and equal in dignity and rights. The Constitution of the Republic of Korea also stipulates that all citizens shall be assured of human worth and dignity and have the right to the pursuit of happiness. The Constitution of the Republic of Korea stipulates human rights in the name of basic human rights*, to inform they are rights that all citizens should be guaranteed, and emphasizes that human dignity should be rightfully guaranteed even if the Constitution does not elaborate on the details of individual rights.

*Natural Rights Natural rights are the rights

inherent to human beings due to their nature as human

beings possessed throughout their lives.

* The Universal Declaration of Human Rights

It is a universal declaration on human rights adopted by

the United Nations General Assemblyin1948and

stipulates that the fundamental rights of every human being

must be respected.

* Basic Human Rights Refers to the basic human right of citizens that are guaranteed

by the Constitution.

The social systems and various technologies and objects created by human beings are used to implement the human rights that human beings deserve to enjoy. However, just as light accompanies shadows, these technologies and systems created to protect human rights sometimes violate them. In particular, AI technology is transforming human life, but at the same time, cases of AI violating human rights are increasing accordingly. Therefore, we need to consider this issue from the ethical perspective.

AI technology is affecting human life throughout society, including medical care, labor, employment, welfare, and public administration. Engineers and companies developing AI systems assert that AI enhances national competitiveness and improves quality of life. However, in terms of guaranteeing basic human rights, AI technology is causing many ethical issues related to the right to personal information and private life, and the right not to be discriminated against, etc. Examples of human rights issues caused by the use of AI technology

‌Face recognition technology is recognized as being very effective for certain purposes such as criminal investigations and security enhancement at airports. The shopping industry uses AI to prevent theft and analyze customer responses to come up with effective marketing strategies. However, AI identified the wrong person as a criminal to be detained, revealing its racial bias. 2 ‌In the process of using AI to determine new hires or find cases of unfair receipt of welfare subsidies, some people suffered from unfair treatment due to technical problems.

‌ Face recognition technology is used for various purposes such as strengthening security, collecting information, and investigating customer responses, but it is also triggering controversy related to the violations of human rights violations in the relevant fields.

Thinkaboutwhyit is important to build

trustworthy AI to protect human rights.

We need to take a closer look at the AI systems’ potential discrimination against people based on sex, race, color, ethnic or social origin, genetic characteristics, language, religious or political beliefs, minority, property, birth, disability, age, or sexual orientation. A thorough ethical check should be conducted in advance of problems that AI technology may cause due to technical errors or inherent bias in AI training data. In the stage of developing, operating, and using AI systems, measures to prevent risks, such as securing various security mechanisms or safeguards to protect personal information, should be taken. The provision of services tailored to each individual requires the maximum amount of personal information. However, if we fail to adequately prepare technical measures and legal regulations to protect personal information in this process, the basic rights that individuals should be guaranteed by design may be violated.

‌Thorough ethical reviews on bias in training ‌Technical devices and legal regulations to data should be performed.

protect personal information are required.

People who deliberately abuse AI systems cause great harm to society. Furthermore, this can lead to serious violation of basic human rights of individuals and undermined values of democracy. For example, deepfakes based on AI technology manipulate public sentiments by making up fake news. In addition, the echo chamber effect produced by the recommendation algorithm amplifies or reinforces specific thoughts by continuously exposing people to only information supporting their preexisting beliefs.

✽

✽

We will learn problem of bias more about the in the training data in the ‘Data Management’

lesson!

* Deepfake This te rmis co inedby combining the words ‘Deep Learning’ and ‘Fake’, and refers to AI-based human image composition technology.

* Echo chamber effect. An echo chamber means an artificial room that gives a reverberating effect to recordings. The ‘echo chamber effect’ is a metaphor to describe a phenomenon in which the thoughts of a certain group are repeated and reinforced like an echo, preferring only certain media due to an AI algorithm that recommends information with similar beliefs or values to its own.

In other words, when people’s biases are strengthened by AI algorithms, it is difficult to accept new information or ideas, preventing the creation of a solid democratic foundation for productive conversations to solve problems. Since people have a psychological tendency to access only familiar or preferred information, excessive reliance on AI algorithms can worsen biased thoughts and amplify social conflict.

5 protection

Ways to develop and use AI in favor of human rights

First, AI technology should be developed and operated to prevent any violation of the rights protected by national laws, including the freedom of expression through media, publication, assembly, and association. Second, AI should guarantee the dignity and values of humans that must be respected for its existence itself, prevent anyone from being discriminated against for unfair reasons or given circumstances or conditions, and allow matters regarding personal information to be independently determined by the relevant individuals. Third, the process of developing and using AI systems should be disclosed so that the general public can understand, and reasonable procedures should be prepared to remedy the infringed rights in the event of human rights violations. AI affects not only the basic conditions for individuals to live with dignity, but also overall society, spanning from a daily life of each member to political beliefs and economic wellbeing.

Designing guidelines for human rights

In the process of developing, using, and evaluating AI, human rights should not be compromised. Institutional and technical guidelines for human rights should be established to protect human dignity and basic human rights through constructive dialogue at all levels of society.

Necessary measures to minimize unexpected social problems and ethical controversies arising from AI development

‌There is a concern that the communications among various groups might be reduced, and individuals’ autonomy and critical abilities can be weakened due to the echo chamber effect.

Company C has been put in the center of the controversy over the violation of privacy after it had acquired the patent for its first face recognition technology in Country P. This is because Company C did not obtain permission from the relevant parties when collecting images from social media platforms. On the grounds of the law, which authorizes the use of publicly available materials, Company C built a database of over 10 billion photos and developed a technology to identify people’s faces and find people with similar faces using a face search engine with a machine learning algorithm. Company C’s face recognition technology is already in use in more than 2,400 state agencies to identify suspects. A human rights group with negative views on face recognition technology criticizes Company C’s technology for using the image database without obtaining owners’ consent, claiming that this has infringed privacy and constitutes a violation of international human rights law. Moreover, the group argues that face recognition technology could have a negative impact on minority ethnic groups. They are concerned that face recognition technology might tend to mistake women and minorities for criminals due to the relatively high proportion of white men used when training the AI algorithm so that it was less accurate in identifying men of color. In fact, the police department of Country P arrested and detained an innocent citizen as a criminal by using face recognition technology for criminal investigations, and as a result, was sued for unfair infringement of personal liberty. The civil liberties union of Country P filed a suit against Company C on a charge of violating the personal information protection act on biometric information, resulting in Company C stopping the sale of the technology to private companies. Country Q ordered Company C to delete the database containing any information on its citizen, and many institutions joined in filing a lawsuit against Company C. Country R’s national watchdog organization on data imposed a fine of $23 million on Company C for its use of data and images scraped from the internet and biometric information for face recognition. <Source> AI Times, Feb. 02, 2022

What could be the problem?

If the government discloses data collected for public purpose to an unspecified number of people or private companies at no cost, the right to protect personal information may be violated and controversy over privacy violation may be sparked. It is also controversial for private companies to pursue profits by using the provided data for purposes beyond the scope of scientific research. Furthermore, the government’s focus on implementing policies for new technology may cause unequal access to information.

Find the cause of the problem and briefly describe it.

Country A disclosed 480 million pieces of data of 170 types in 8 areas including health care, vision, safety, and autonomous driving for its project called “Promotion of the Development and Use of Public Data.” However, since most of the collected and shared data are collected in real life environments, except for some data artificially produced with the consent of the individual, there are concerns that personal information or sensitive information of citizens included in the large amount of data for AI training might be leaked. In particular, the health care data or data collected from the digital care project for vulnerable groups in the medical field contains sensitive information about actual health, such as disease information, and there is a high risk of infringement of privacy. Some argue that it is also unreasonable to allow companies to receive sensitive information from the public government free of charge and use it to generate profits. The government processes and distributes data related to autonomous driving and safety that is collected from CCTVS in public areas such as subway stations, roads, and inside buses. Personal information such as a person’s face and a license plate included in those image information has been de-identified through mosaicing or blurring. However, in the future, we might encounter a problem related to personal information protection, as faces can be detected or recognized from the same image information thanks to the advanced deep learning technology. Above all, it is a violation of the Personal Information Protection Act to disclose sample data, etc. to an unspecified number of people or to provide pseudonymous information to private companies to be used for purposes other than scientific research.

<Source> National Human Rights Commission of the Republic of Korea, “Study on Human Rights Guidelines in the Development and Use of Artificial Intelligence (Final Report)”

If the government discloses data collected for public purposes to an unspecified number of people or private companies at no cost, the right to protect personal information may be violated and controversy over privacy violation may be sparked. It is also problematic for private companies to pursue profits by using the provided data for purposes beyond the scope of scientific research. Furthermore, the government’s focus on implementing policies for new technology may cause unequal access to information.

Find the cause of the problem and briefly describe it.

Explore the relationship between AI systems and guaranteeing basic human rights

Select and analyze one of the following problematic cases related to basic human rights that we

might face in the process of using AI technology.

Problematic cases

Algorithms used to recruit new employees generally revealed a preference for men over women.

Online chatbots showed racial ideologies only within a few hours.

Machine translation exhibited a gender bias.

The face recognition system correctly detected the faces of white males, but not those of black 4 females.

Category

Problematic case chosen by the group

Overview of the case

Type of violated basic human rights

The technical aspects of the AI system that caused the infringement of basic human rights

Details

For types of basic human rights, refer to the applicable provisions of the Constitution

the Universal Declaration of Human Rights, or human rights-

related sections in ethics and social studies textbooks.

Solution

Objection

Rebuttal

on the following topic.

‌Suggest and discuss the solutions to the violations of basic

human rights explored earlier in groups from the technical

institutional, and ethical aspects, respectively.

Technical aspect

Institutional aspect

‌Based on the discussion results, write a 300-word essay

How should AI systems be developed and utilized to protect human rights?

Anticipate other groups' Safeguarding Human Rights objections to your group’s solution, and think about how to respond to

those objections again!

Ethical aspect

Write an essay by dividing sections into the

technical, institutional and ethical aspects.

Make a human rights impact assessment system to protect human rights

‌Investigate the purpose and items of the Environmental Impact Assessment System, and create a Human Rights Impact Assessment System that can be applied in the process of AI development.

Based on the environmental impact assessment system conducted prior to a development project, discuss setting specific purpose and necessary assessment items to protect human rights when developing AI.

Category Environmental Impact Assessment System Human Rights Impact Assessment System

Purpose

1. Air

3. Land

Assessm- ent item

4. Natural ecology

6. Society and economy

intended to predict and analyze environmental The Environmental Impact Assessment System is impacts that may occur in the process of implementing various policies, plans, or development projects and to prepare measures to minimize harmful effects, thereby pursuing the conservation and development; second, seek eco- following aims. First, achieve a harmony between promote the health and well-being of citizens. friendly and sustainable development; and third

How bad of an odor could the development cause? 1.

How much could the development destroy the 2. Water environment marine environment?

How might the development change the surrounding 3. geographical features?

How much of a wetland conservation area could be

lost due to the development?

How much noise could be generated in the 5. Life environment development process?

How might the development change the industrial 6. structure?

‌Discuss and share the reasons for the assessment

items of the human rights impact assessment

system selected by each group.

Assessment item

Before presenting reasons for the need for evaluation items of the Human Rights Impact Assessment System, predict and Safeguarding Human Rights

discuss potential human rights violations in relation to the relevant assessment item!

Why it is required

‌Have an open discussion on the ways to prevent human rights violations with respect to the

You can understand that basic human rights may be violated in the process of developing and using AI.

group’s chosen item from the Human Rights Impact Assessment System.

You can propose solutions to the problem of infringement of basic human rights that occurs while developing and using AI from the technical, institutional, and ethical aspects.

You can predict the human rights violation situation and logically present the basis for the assessment items to be included in the Human Rights Impact Assessment System

required for the AI development process.

Advanced Problem Solving Activity Problem Solving Activity

Ethical review of AI-based social credit systems

‌Read the following article about the AI-based social credit system in Country A, compare <Perspectives 1 and 2> in terms of human rights protection, and write your own thoughts.

To put it simply, Country A’s ‘AI-based social credit system’ uses big data to score people’s behavior to give benefits to people with high scores and disadvantages to those with low scores!

Arguments for and against the AI-based social credit system

Since 2014, Country A has been introducing and building an “AI-based social credit system” with the aim of expanding the system nationwide by 2020. This issue is of great significance in that it is related to the national information management at the state level in the era of big data. Country A already initiated its discussions on building a foundation for the social credit system back in 2007. In May 2014, Country A announced the policy to introduce a social credit rating system in which each person and company is individually evaluated and graded by 2020. Country A specified the purpose of introducing this system as raising awareness among members of society, establishing a desirable trust society, promoting social development, etc. However, the ulterior intention of Country A is to use the trust among members of society as a medium to enable the government to make and manage a database containing information on the appraisal of loan repayment ability and individual’s potential social contribution, judgment of individuals’ tendencies, and surveillance of offenses. Country A has introduced and implemented the social credit system after designating and announcing a total of 28 pilot cities for building a social credit system on two occasions in January 2018 and August 2019. Each pilot city has selected detailed evaluation items for social credit evaluation, calculated scores based on such items, and imposed penalties relating to the use of transportation service use and social welfare benefits on individuals with low credit ratings. <Source> Seung-eun Lee, “Comparison of the opinions of Students A and B on Country A’s social credit system” (April 2019)

Perspective 1 Traditionally, people in Country Perspective 1 A have not trusted each other easily. Those who have close relationships with people in power are more respected than those who are skilled or hard-working. Those with close ties to the powerful can also avoid The data would include information about all parts of our lives, punishment even if they break the law. including our finances and political beliefs. I know someone who If Country A can create a good social credit system, has bad credit, but he is not a bad person. He used to have a it can become a more trustworthy society where successful business, but he was hit by the economic recession and social systems are run in a fair and transparent way. went into debt. Therefore, I believe that citizens should support If the social credit system is in place, it will be very difficult for the government’s creation and use of a database of people who make mistakes or fail to get economic and cultural people’s biometric and financial information to build benefits. Isn’t it the government’s job to help its people and give a social credit system. them a second chance when they fail?

The social credit system that Country A plans to use has many problems. The government wants to create a database of people’s information, rate the information, and use the results as credit scores. This would create a controlled society.

Category

Perspective 1

Perspective 2

Your opinion

Assertion

Grounds

‌Decide whether you are in favor of or against the introduction of the AI-based social credit system in

Support

Oppose

Country A, and write down the reasons.

Who would support the introduction of the system?

Reasons for the assumption

Who would oppose the introduction of the system?

Reasons for the assumption

‌Express your thoughts on the introduction of the AI-based social credit system with a 4-column cartoon.

You can share your opinion on the introduction of the AI-based social credit system with supporting grounds.

You can compare the advantages and disadvantages of AI-based social credit systems.

Company S, an artificial intelligence development company, agreed to provide data National Defense in Country P last February. classification technology to the Ministry of intelligent drones with obstacle avoidance Company S, a company well known for its features, also recently signed a contract to supply drones worth $100 million to the army also a party to a contract to the amount of $1 of Country P. Another company, Company A, is Defense of Country P in February this year. It is billion executed with the Ministry of National build a drone defense system. a contract for a 10-year project to develop and

AI Systems Helping the Defense Industry

Military training using drones

world, with a scale of $151 billion a year, is also seeking industrial innovation. This is because, in addition to As technology changes the wartime landscape, Country P, the country with the largest defense spending in the expensive fighter jets and tanks, the importance of advanced weapons using AI, big data, and drones and the command and control system that connects them is growing. To meet these needs, the Ministry of National Defense of Country P is collaborating with big tech companies such as Companies M or G, and investing in startups to develop technologies that boost the efficiency of war efforts and save training costs. The economic magazine Fortune stated, “In the past, private companies used state-of-the-art technologies developed by government agencies such as the Ministry of Defense or the National Aeronautics and Space Administration. But now, in contrast, the government is creating funds across all ministries to introduce technologies developed by private companies.” Due to the recent war, inside and outside the Ministry of National Defense of P, the opinion that ‘the government and AI development companies must join forces for national security’ is gaining more strength. In the past, big tech companies were reluctant to work with the Ministry or engage in war-related businesses. As a typical example, in 2018, Company G worked on a joint project for drone strikes using AI technology with opposition. But now, the defense industry is recognized as a pivotal asset that protects people’s lives. the Ministry of National Defense of Country P, but it had to withdraw from the project due to severe internal

<Source> The Chosun Ilbo, Sep. 08, 2022

In recent wars, a range of technologies using AI are being deployed. Examples include deepfakes that reduce the morale of enemy soldiers through manipulated videos, machine learning

technology to determine whether war videos are synthesized, drone attack technology that minimizes troop damage, and technology to analyze data posted on social media about the

deployment of other forces or attack situations. In a peaceful time without war, public opinion against AI-related technologies used for military

purposes is strong. But when a real war breaks out, such criticism is quickly forgotten. When World War II broke out in the past, scientists and engineers agreed to develop an atomic bomb for the

victory of the Allied Forces, but after seeing the terrible situation caused by the atomic bombing, they regretted their choice and participated in the anti-war movement. Considering this historical

case, AI technology may be following a similar path.

1. For each of the main concepts presented, write down at least 5 words that come to mind in relation to ‘Safeguarding Human Rights.’

Concept

Face recognition technology

Surveillance society

Social credit system

Words that come to mind in relation to Safeguarding Human Rights

2. Write down three things that you thought are most impressive or important while learning about ‘Safeguarding Human Rights,’ one of the requirements of AI ethics.

3. Write down three things that you wanted to further explore while learning about ‘safeguarding for human rights,’ one of the requirements of AI ethics.

Safeguarding Human Rights

You can find and present the details of human rights protection required in the era of AI from the Constitution and the Universal Declaration of Human Rights.

You can anticipate various situations in which human rights may be violated when developing or using AI.

You can propose solutions to the problem of infringement of human rights which occurs while developing and using AI from the technical, institutional, and ethical aspects.

Ⅱ

Protecting Privacy

You will be able to understand the meaning and importance of privacy protection in the development and use of AI. You will be able to suggest measures to protect privacy and minimize the misuse and abuse of personal information in the development and use of AI in the personal, social, institutional, and technical aspects.

Name

Food

Phone number

You will be able to come up with measures to protect personal information in the era of AI by understanding the concept of privacy protection and investigating cases where privacy is violated or personal information is misused by AI.

I should think about ways to protect privacy.

Religion

Skin color

Religion

Violation record

‌AI using big data can address and help with various tasks in our daily lives, but there are cases in which privacy is not properly protected in the process of using AI. What measures can be taken to protect privacy?

It is expected that there will be traffic jams due to a major traffic accident that occurred on Olympic Road this morning.

November 24, 20XX, evening

I heard a lot of criticism from my boss at the company today. He got mad at me, but he didn’t even understand my report correctly. It’s so annoying.

November 2, 20XX, morning

Saturday, December 13, 20XX

It is a chilly morning. According to the smartwatch, your current body temperature is 37.8 degrees, and your heart rate is 1.3 times faster than usual. Hi ★★, Could you please order a cappuccino at ◇◇ café near the office for 8:30?

Your voice sounds different than As you requested

I have sent to your

usual. You sound quite sad today. email the audio files

of our conversations over the past year.

Hi ★★, Can you search for local hospitals open on Saturday morning?

December 31, 20XX, evening

Thanks ★★. Oh, wait. The file seems a bit strange. Judging from the transcript, it’s an audio file of a stranger with the same name as myself.

What is privacy?

Meaning of privacy protection

‌ ‌Examine the concept of Privacy protection, one of the 10 key requirements.

Privacy refers to the right not to be interfered with by others regarding matters related to an individual’s personal life. This right is extremely important in relation to human rights in modern society and a key right protecting a person’s individuality. We all are unique individuals. An attitude of respect for the unique identity of every individual is part of the necessary ethics required for everyone to lead a decent life. People’s desire for privacy varies. Some people are not uncomfortable posting photos of their faces on social media that can be viewed by strangers, but others are extremely reluctant to reveal aspects of their daily life even to their closest friends. Different tastes and preferences for privacy is an important factor that distinguishes one individual from another.

In order to protect each other’s privacy while living with others, we must have an attitude of respecting each individual’s different characteristics. Privacy originally meant ‘the right to be alone.’ Therefore, we should respect other people’s egos to the extent that the privacy of others is not infringed upon or the value of universal human rights* is not compromised. Ethics for Protecting Privacy are receiving more and more attention. However, as information communication technology including the internet develops and cyberspace expands with the advent of social network services, the areas where privacy should be protected are expanding, and individuals, companies, and countries are revealing divergent opinions about privacy protection. In particular, as AI systems are incorporated into various technologies and used in daily life, we are witnessing increasing ethical issues related to privacy protection.

Relationship between the development of AI technology and privacy protection

With the development of AI technology, a new technical landscape providing services that help maintain social relationships or are tailored to each individual is emerging. Customization using AI takes the form of providing specialized services to users after analyzing and inferencing the interests, preferences and tendencies of users by processing their online behavioral information*. One of the factors that contributed the most to increasing the value of AI is personal information*.

AI technology that has beenusedtomake individual life convenient may actually violate individual privacy. Let’s think about the duality of AI technology!

* Universal Human Rights It refers to the human rights that all people should enjoy simply on the grounds that they are human beings. It is a right that every human can have regardless of nationality, religion, gender, age, occupation, etc. * Online Behavioral Infor- mation Information on online user activities that is used to identify and analyze users’ interests, preferences, and tendencies, including website visit history, app usage history, and purchase and search history. * Personal Information Personal information is information about a living

individual that enables personal identification, such as name, resident registration number, and image, and also includes information that cannot identify a specific individual by itself, but can be easily combined with other information to enable individual identification.

Great opportunity to shop to your taste!

It is very convenient, because AI Your skin is already needs moisture aware of my preferences. protection!

This is because the personal information of a great number of people is collected and used as big data. Big data is widely used in establishing policies of public institutions or in the marketing process of companies, and its usage is further increased as it is combined with a range of different AI technologies. However, ethical issues related to privacy may be brought about in the process of using personal information through AI technology by companies, countries, or individuals. An individual’s sensitive information may be leaked and the his/her right to privacy and freedom may be violated. Furthermore, someone may trigger another social problem by misusing such information.

✽

* Sensitive Information Protecting Privacy Any information including ideologies, beliefs, admission to or withdrawal from a trade union or political party, political opinions, health, sex life, and other personal information that is likely to markedly threaten the privacy of any information subject

We have found out today is your birthday based on the information from other sites!

How does it know my information?

I heard you are suffering from constipation!

Advantages and disadvantages when personal information is used for customized services using AI technology

may be exposed to or misused by others, resulting in personal as well as economic damage.

Case

When developing AI, various types of personal information are collected and used, and in this process, privacy infringement-related problems may arise unintentionally. In order to address these problems, we need to enhance awareness of the potential violation of the freedom of privacy in the process of developing and operating AI systems, and eliminate risk factors in those processes.

In using personal information to develop and operate an AI system, information on an individual’s private life

‌When an AI system using CCTV is developed and operated to prevent crimes, individuals are subject to surveillance, or their privacy is unfairly damaged due to technical errors.

Case

Privacy protection-related issues arising in the process of using AI systems

In the course of developing an AI system, privacy may be violated due to poor data management.

‌De-identification is not properly performed on personal information to be provided companies, or such information is used to government agencies or private outside the intended purpose of provision.

De-identification process of personal information for privacy protection

Step 1

Preliminary review

Personal information (Identifiable

information)

In this era of big data, appropriate use of personal information helps to not only increase the efficiency of society as a whole, but also to generate economic benefits. However, when building an AI system using personal information, de- identification* measures should be conducted for privacy protection.

Step 2

Step 3

De-identification

Adequacy evaluation

Non-identification adequacy

De-identification (Removal of personally identifiable elements)

evaluation

No (inadequate)

Information that can be used alone to identify a specific individual (including information that can be easily used with other information to identify a specific individual).

* De-identification It refers to a process of deleting or replacing a

pa rtorallofpersonal information by means of data

reduction, pseudonymization, aggregation, data suppression

data masking, etc. so that the information cannot be used to

identify a specific individual even when combined with

other information.

Follow-up management

Yes (adequate) Non-identifiable

information

Presumed to be non-personal information (however, necessary management measures should be taken to ensure that it is not used with other information for re-identification).

<Source> The Guidelines on De-identification of Personal Information, 2016, The Ministry of Public Administration and Security

If personal information undergoes the de-identification step, it turns into pseudonymized information and anonymous information. Pseudonymization of information refers to a method of processing personal information so that it cannot identify a specific individual without additional information by deleting or replacing a part or all of the personal information. Anonymous information is information that cannot re- identify an individual even when used together with additional information.

If it is closer to personal information The value for use

The possibility of privacy violation

of information

If it is closer to anonymous Information

The value for use of information

The possibility of privacy violation

Importance of the Right to Self- determination of Personal Information

J.S. Mill, a British liberal thinker, stated in his work On Liberty, “Over himself, over his own body and mind, the individual is sovereign.” This sentence can be reinterpreted to mean “In the era of AI, everyone has the right to control his/her personal information.” The right to control personal information is also called the “right to self-determination of personal information.” The ‘right to self-determination of personal information,’ which refers to the information subject’s right to make a decision on his/her own concerning the disclosure and use of his/her personal information, can be justified on the grounds of basic human rights stipulated in the Constitution. That is, the right to self-determination of personal information can be guaranteed on the basis of the right to pursue happiness with human dignity and value and the right not to be infringed on the secrecy and freedom of privacy as prescribed in the Constitution. This right allows everyone to decide for himself/ herself when, to whom, and to what extent information about him/her is made available and used.

Government, private companies and AI systems

I will decide to allow the use of this part of my

personal information only over one year

I need your personal information.

Personal information

Developer

I will control and make decisions

The discussion on the ‘right to self-determination of personal information’ has grown in about my personal information on my importance with the advent of the information age. Together with values such as the own. ‘right to know’ and the ‘freedom of expression,’ this right is triggering various ethical debates. As science and technology become more advanced, it is highly likely that the current personal information consent system will not be applied in practice. This is because in the advanced form of the Internet of Things (IOT) system, in which the collection and use of information is carried out without human intervention, it can be very difficult to obtain prior consent from each individual for the collection and use of personal information. Therefore, individuals, companies, and countries are required to draw a social consensus on the ways to efficiently develop and utilize AI technology while protecting privacy.

Chatbot that gives you the feeling of having a conversation with a human

The relationship-oriented chatbot ◯◯ from Company S suspended its service in January 2021, a month after its launch due to ethical and legal controversies. The ◯◯ chatbot service developed based on deep learning technology had gained tremendous popularity as a dynamic service that replicates the experience of having a conversation with a real human, unlike previous chatbots. However, when the user prompted a specific conversation, the chatbot made a series of hateful and discriminatory remarks. Even in the process of sharing development records of chatbot ◯◯ on an open source sharing platform, personal information was provided without being de-identified, further amplifying the controversy related to personal information. The Personal Information Protection Commission (the Commission) imposed a corrective order and a fine for collecting personal information of children under the age of 14 without obtaining consent from their legal representatives and for using sentences from K-talk conversations to train and operate the chatbot beyond the purpose of information collection. After the discontinuation of the service, the company felt heavy responsibility for the social controversy and is putting more efforts into personal information protection including making AI chatbot code of ethics and self- checklists, examining the safety of conversation, and separately managing training data and the database for Chatbot ◯ ◯’s responses. The reason Chatbot ◯ ◯ became so popular was because it mimicked the feeling of having a conversation with a real person. As AI technology continues to develop, this type of chatbot will become more technologically advanced, leading to the possibility of even more ethical and legal controversies relating to personal information arising at any time.

What if the following problem occurs?

Through an investigation, the Commission verified that Company S used K-Talk conversations collected from its app service for the development and operation of Chatbot ◯ ◯ from February 2020 to January 2021. With respect to the use of users’ K-Talk conversations by Company S for its development and operation of Chatbot ◯ ◯ service, the Commission determined that even though Company S specified the purpose of collecting information as ‘development of new services’ and then obtained consent from users, users could not have predicted that their information would be used for the development and operation of Chatbot ◯ ◯. In addition, regarding the developers who posted conversational texts containing personal information (22 names, 34 regional names, gender and relationship with conversation partner) on the website for code-sharing and collaboration, the Commission judged that they violated paragraph 2 of Article 28-2 (Processing of pseudonymized information, etc.) of the Personal Information Protection Act on the grounds that they provided an unspecified number of people with pseudonymized information containing the information that could be used to identify a specific individual. In addition to personal information issues, the chatbot’s hateful and discriminatory remarks caused ethical controversies.

<Source> Press release from the Personal Information Protection Commission, Apr. 28, 2021

Find and briefly describe the reason why the problem occured.

Efficiency and risk of the use of biometric database

In June 2022, Resident ★s of Area A in Country H boarded a plane to participate in a protest against a bank located in Area B. Resident ★ had been depositing money in a small bank in Area B, but the bank suffered financial difficulties and suspended paying deposits from last April. Therefore, Resident ★ and two other depositors decided to visit Area B to complain to the competent financial authority. However, within a few hours of arriving in Area B, the COVID-19 health code displayed on Resident ★’s mobile app suddenly changed from green to red, indicating that he needed to be quarantined, and as a result, Mr. Kim had to return to Area A. Resident ★, who was not infected with COVID-19, complained to the quarantine authority, but did not receive any response. Resident ★ was not the only person who encountered this unusual situation. Many of the victims of the bank default saw their health codes turning red on that day. The account holders argued that the authority in charge of managing the health code in Area B, manipulated the code to prevent the protesters’ visit in advance. A citizen of Country H said, “This case demonstrated that personal information collected under the pretext of quarantine can be used by the government for monitoring and control of individuals at any time.” There is also a growing criticism over Country H’s collection of extensive personal information with advanced technology and using the information to control the state affairs. Country H is running an AI service that collects voices with surveillance cameras, analyzes the voiceprint, builds an iris database, and collects the Y chromosomes of males to track criminals. Furthermore, Country H is also operating a project to identify people who install the Language C dictionary app on their mobile phones to monitor Minority Group C. With 2.5 billion images of faces collected through security cameras, concerns about AI technology that monitors residents are spreading.

<Source> The Chosun Ilbo, Sep. 06, 2023

Each country around the world makes identification cards to administratively manage their residents. On these identification cards, personal information can be indicated in the form of numbers, such as a resident registration number or social security number, but biometric information, such as a photo or fingerprint, may also be recorded. Recently, biometric information such as a face or fingerprint is used to identify an individual more conveniently in an airport, port, or public building. As AI technology advances, the domain of biometric information that can identify individuals, such as face, iris, fingerprint, vein, and retina, is expanding, and DNA research is also gaining momentum these days. However, if biometric information is leaked or misused in the process of using contactless services during the COVID-19 pandemic or other similar situations, infringement of personal information may occur. Also, other problems may arise while biometric information is shared among administrative agencies or countries for the purpose of preventing terrorism or crime.

Briefly describe what problems might arise when biometric information is used.

데이터 관리

Explore the ethical controversy over AI speakers

‌Identify three positive uses for AI speakers and write them in the chart below. You may use the reference words in the box to help you.

Reference words

Searching information, finding directions, playing music, the visually impaired, the elderly, financial payment, hospital emergency room, purchasing products, emotional connection, friends, self-driving car, Internet of Things, etc.

Positive use 2 case

‌Referring to the following words, find ethical controversies that may arise from positive use cases

of AI speakers, and select one of them to discuss its solutions.

Reference words

Defamation, deepfake, racism, personal information leak, Big Brother, eavesdropping, financial fraud, security, technical error, infringement of privacy, de-identification, user’s characteristics, addictiveness, political orientation, recommendation of its own website, etc.

Ethical controversy

Ethical controversy selected by the group

Solution

Objection

Rebuttal

Anticipate other groups’ objections to your group’s solution, and think about how to respond to those objections again!

Delve into the issues on AI technology and privacy through mythology

‌Read the story of Odysseus and the Sirens adapted for the use of personal information

in the era of AI, and then find and write cases suitable for the following questions.

Since humans are the developers of AI technology, we need to understand our ways of thinking. Ancient mythology embodies a range of different values.

Therefore, analyzing the symbolic story of mythology contributes to the desirable development of AI technology!

Sirens are witches who appear in Homer’s ‘The Adventures of Odysseus.’ Living on an island in the Mediterranean Sea, they seduce passing sailors with their enchanting songs and eventually cause their death by tricking them into shipwrecking their boats on rocky shores. When Odysseus passes near the island of the Sirens, he commands the sailors to plug their ears with wax beforehand. But instead of plugging his ears with wax, Odysseus instructs his sailors to tie himself to the mast. He earnestly asks the sailors to tie the rope tighter and never let him loose if he begged to be untied by the sound of the song. As expected, when the ship of Odysseus and his crew passes by the witches’ island, the Sirens began to sing bewitching songs. Odysseus is completely captivated by the songs and struggles from the mast to get closer to the shore where sirens live. The sailors gird Odysseus tighter as they are instructed in prior, and eventually the ship safely passes the Sirens’ Island and continues on a smooth voyage.

Sirens’ voices

The rope that is used to tie Odysseus to the mast

Reconstruct the story of Odysseus and the Sirens as follows.

Plugging sailors’ ears with wax

Personal information used for developing

AI technology

Losing a life lured by the Sirens’ voices

Social regulations to prevent misuse of

personal information

John William Waterhouse, Odysseus and the Sirens, 1891.

Excluded from the benefits of using personal information

Mental and material damage caused by

misuse of personal information

The sailors who plugged their ears with wax did not hear the Sirens’ sweet voices at all for fear of using their autonomy. In contrast, Odysseus prepared for the worst scenario and enjoyed the Sirens’ voices to the fullest by exercising restraint. Based on the above mythological story, we can learn a lesson that in order to properly use optimized technology in the era of AI, we need to secure the maximum quality data, while putting in place the minimum stringent restrictions at the same time.

Question

Personal information used for developing AI technology (Sirens’ voices)

Benefitting from AI technology using personal information (Odysseus enjoying the sweet voice of the Sirens)

Damage caused by improper use of personal information in developing AI technology (losing a life lured by the Sirens’ voices)

Social regulations to prevent the improper use of personal information in developing AI technology (the rope tying Odysseus)

Have a free discussion on the application of the symbols in the mythological story of the Sirens and Odysseus to the use of personal information in the AI era!

Corresponding case in the era of AI

‌

Find ways to use AI that contributes to privacy protection

Seek ways to prevent violation of privacy by AI from the individual, social institutional, and technological aspects.

Individual

Based on investigated cases in Activity 2-1

Have a group discussion examining the different solutions for each selected aspect and critically analyzing the advantages and disadvantages of each solution.

Social institution

Technology

‌Have a practical exercise of de-identifications for personal information by referring to the

Pseudonymization

Aggregation

Data reduction

Data suppression

Data masking

Category

Pseudonymization

Aggregation

Data reduction

Data suppression

Data masking

You can suggest ways to protect privacy while using AI in relation to the personal, social, institutional, and technical aspects.

examples below for the 5 types of de-identification measures.

Examples of de-identification

How-to

Replace the value of a personally identifiable Ron Weasley, old, studying at Harvard 25 years Harry Potter, 25 years old

variable with other values University

Use the sum or average of the data

Student C’s height: 160 cm 172 cm

Remove the value of a personally identifiable Harry Potter, 981212-323456, 26 years old, Male, living in

living in Boston

variable.

Replace explicit data values with values for broad ranges Ron Weasley, living in Boston 25 years old, Ron Weasley, 20s

Mask the values of personally identifiable variables to make Ron Weasley, studying at Harvard University

them invisible

Personal information

Harry Potter, 45 years old, studying at Stanford University

Harry Potter’s weight: 71 kg Hermione Granger’s weight: 62

kg. Ron Weasley’ weight: 83kg

Ron Weasley, 981122-123456, Living in Boston

Betty Kane, 57 years old, living in New York

Harry Potter, studying at Oxford University

You understand that privacy can be compromised in the process of using AI, and can find examples of potential ethical controversies.

Before de-identification De-identified information

studying at University A

Student A’s height: Student B’s height: 182 cm 174 cm Student height Student height Sum: 516 cm Average

Boston

living in Massachusetts

University ◯◯, ◯◯ Wayne, studying at

98✽✽✽✽-1✽✽✽✽✽✽

De-identification

Explore ethical issues of using self-driving cars

Search the internet and write down the characteristics of each level of driving automation.

Level

Definition

No driving automation

Driver assistance

Partial driving automation

Conditional driving automation

High driving automation

Full driving automation

Characteristics of driving

Read the following article and learn about the ethical issues that may arise from operating self-driving cars.

Find answers to questions by searching for various references such as newspapers!

Protecting Privacy

Personal information protection, a prerequisite for vitalizing the self-driving car industry

Amidst a fierce race to develop self-driving cars in the domestic and foreign automobile industries, it is increasingly argued that institutional improvements, such as taking measures to protect driver’s personal information collected while driving self-driving cars, should be carried out during the process. According to this view, if personal information collected in self-driving cars is not securely protected and is misused, there is a risk of damage to the driver as well as secondary human and material damages from use of the data. For this reason, people point out that the safe protection and use of personal and location information is a prerequisite for the prosperity of the automobile industry. Self-driving cars collect a wealth of information such as driver’s personal information and location information to control all safety functions while driving. Self-driving cars themselves can be considered to be forms of “big data.” Therefore, developed countries such as the United States and the United Kingdom are taking the lead in enacting laws or guidelines to protect personal information related to self-driving cars. In September 2017, the U.S. House of Representatives unanimously passed the Self Drive Act, which regulates cyber security and privacy issues related to the use of self-driving cars. This act, which applies equally to all 50 states in the United States, stipulates that drivers shall be allowed to identify the types of data collected and stored when driving self-driving cars and to refuse the collection of their personal information, such as personal biometric information and behavioral information. The act also states that self-driving cars must be equipped with security systems that can defend against cyber attacks, while manufacturers must record and share security-related programs and evaluations. <the rest omitted>

Questions about ethical issues ‌What personal information is required for the safe operation of self-driving cars?

<Source> The Law Times, Nov. 16, 2017

Answer

Propose solutions to ethical issues arising from the operation of self-driving cars

Driving automation level 3 (conditional driving automation) Mr. A is driving a car

The AI is driving autonomously while checking the pupils of Mr. A who left work at dawn.

Mr. B is driving a car (conditional driving automation) in the driving automation level 3

After checking your health information, I recommend that you take a break at the next rest area.

such as the driver’s Health information heart rate is checked through sensors installed on the steering wheel.

Driver’s personal information

‌In the chart below list personal information that might be used in a self-driving car system in the column on the left. Then, look at the following cartoon and write down why this particular personal information might be useful in the column on the right.

Driving automation level 3 (conditional driving automation) Mr. A is driving a car

AI detects risk factors at the construction site during autonomous driving and sends a guidance signal to Mr. A that he needs to take

the wheel in person based on the data on the size of Mr. A’s pupils measured on a daily basis.

information

Based on the article andthein cartoon in , think of the reasons and write them down!

Activity ❶

Activity ❷

Hold the wheel!

How does it know my exact health condition and send relevant

Hacking big data related to driver’s health Spamming

advertisements?

emails promoting health-related products

Spam

Why the driver’s personal information is necessary

‌Think about and write down the ethical issues that might arise when the driver’s personal

Item number

Ethical issues found in our group

Solution

Objection

Rebuttal

You can understand the ethical controversies concerning personal information that might arise from operating self-driving cars.

information is provided to the self-driving system.

Potential ethical issue

‌Choose one of the ethical issues found in the group

present a specific solution to the issue, and discuss it.

You can suggest the solutions to the ethical controversies concerning personal information that might arise from operating self-driving cars.

Anticipate other groups’ objections to your group’s solution, and think about how to respond to

those objections again!

Delete or replace a part or all of information elements that can identify a specific individual.

A set of private life or personal information that you do not want to share with others.

when companies or public institutions intend to A process that is performed against individuals

use private information.

Privacy

Giving consent to use of personal information

De- identification

2. Write down three things that you thought are most impressive or important while learning about ‘privacy protection,’ one of the requirements of AI ethics.

3. Write down three things that you wanted to further explore while learning about ‘privacy protection,’ one of the requirements of AI ethics.

You can distinguish personal information and non-personal information and explain each of them.

Privacy Protection

You can anticipate possible situations in which privacy may be violated when using personal information.

You can describe the personal, social, and technological efforts required to benefit from AI technology while protecting privacy.

Ⅲ

Respecting Diversity

You will be able to understand the concept and importance of respecting diversity in the context of using AI. You will be able to explain the need to resolve discrimination and bias in the development and use of AI systems in relation to the principle of protecting privacy. You will be able to suggest ways to respect diversity in the deployment of AI technologies and services.

You will recognize the need for diversity in every area where AI technology is used and explore measures to ensure the protection of privacy.

‌AI systems might follow and reproduce the stereotypes and prejudices inherent in our current society, resulting in issues related to bias, appropriateness, and fairness. What efforts should be made to prevent these issues?

guys going to What are you I am not any good movies?

sure.

do tomorrow? It is Saturday!

Write down your opinion.

What issues can Somin bring up in the following situation?

I am thinking about going to the movies this Saturday. Can you recommend

optimized and customized the recommended movies. content from algorithms that predict what you like. How about that?

Yeah, but the algorithms only similar movies. It gets recommend all these boring.

personalized content …

I agree. Those tailored content recommended rather makes me by AI algorithms bored.

Well … These days, you can get recommendations for That is a good idea. You can check out

Well. Maybe that is a problem with algorithm-based

I haven’t the algorithms thought about

personalized content.

that way. What do you think, The problem Somin? of algorithm- based

Protecting privacy

Gun??

‌ ‌Examine the concept of protecting privacy, one of the 10 key requirements.

People have different abilities and opinions, and also differ in ethnicities, skin colors, and religions depending on the location of birth or natural environment. Differences* among people are therefore a very natural phenomenon. However, in human history, there have been frequent cases in which differences among individuals based on social culture or institutions have led to discriminations*. What discrimination may occur in the times of AI and why should diversity be respected? It is because if the AI system imitates and reproduces the stereotypes and prejudices accumulated in our society and culture, the development of AI can eventually become the epicenter of new social discrimination and trigger further social conflicts.

Relation between AI and protecting privacy

In today’s society, the elements of the “Fourth Industrial Revolution” such as big data, algorithms, and AI are forming a new culture. Although automation is increasing our convenience in this new environment, we can also face bias issues that discriminate against users as social prejudices are projected onto AI systems. The ethical value of respecting diversity means that i) diversity and representation of users should be reflected in all stages of developing and using AI, ii) bias and discrimination based on individual characteristics such as gender, age, disability, region, race, religion, and country are minimized, and iii) AI should be applied fairly to everyone in daily life. For example, in an experiment in which an AI was used to recognize objects contained in a photo of a person holding a thermometer, the AI recognized a thermometer held by a black person as a gun and yet when the same object was held by an Asian, it was recognized as an electronic device. This occurred because of biased training data using images that showed black people more frequently in the context of guns. The data was therefore biased. It is important to ensure that training data is diverse and unbiased in the development of AI systems.

Thermometer!!

‌Consequences caused by racially biased training data

* Difference and Discrimi- nation

A ‘difference’ between people indicates that the people

are ‘not the same’ in some objective way. Discrimination

consists of treating people unfairly for their ‘being

different’ without accepting their differences.

bias and implies that users are confined to A filter bubble is a term referring to information only the information category filtered to suit their interests. A filter bubble occurs when a and exposes advertisements based on the service provider recommends related content purchase history, etc., and as a result, the user is user’s search history, video viewing history, constantly exposed to similar information. The reason why the filter bubble is an issue is

media literacy may also help individuals to adopt a critical attitude towards the digital content they consume.

AI-based recommendation system and diversity

Today’s AI-based recom- mendationplatform environment is characterized • There is a sharp distinction between information by personalization and producers and consumers. customizat ion unlike • One-way medium whose information consumers are passive and inactive. traditional mass media such as newspapers, radio and television. Big data composed of emotions, sentiments, and information that we express can form various subcultures* that have previously been suppressed in mainstream culture. This can add dynamism and diversity to our society and serve as a stepping stone to create a new culture. At the same time, however, recommendation systems – algorithms that suggest content to us based on our past behavior and interests – can threaten the value of respecting diversity in the area of production and consumption of popular culture.

Mass media

Differences between mass media and the AI-based recommendation platform environment

Respecting Diversity and filter bubble 4

‌Filter bubble

personalized information, they can easily fall into confirmation bias* because that when users only see, hear, and consume they fail to experience information that is outside their taste or preference spectrum. To solve this problem, users can take measures such as regularly deleting personal information such as search history and viewing history in the device they use. It is expected that such a simple measure will help to prevent In the new era of AI, we should strive to protect and develop a democratic users from being trapped in the filter bubble to some extent. information in advertisements and search recommendations. Enhancing our and diverse culture. This may require preventing excessive use of personal

AI-based recommendation platform

• Information producers and consumers can have two-way communications.

• Delivery of information is faster than conventional media.

• The public can serve as producers and consumers at the same time.

* Mainstream Culture and Subculture

Mainstream culture is a culture that has a dominant influence

in a society and is shared by a majority of society members

while subculture refers to a culture shared only by

members of a certain group.

* Confirmation Bias It refers to a cognitive bias that seeks and accepts only

information or evidence that are in line with one’s belief and

ignores other information that contradicts the belief.

Does AI prefer lighter skin?

In July 2016, Beauty.AI, a global beauty contest where AI robots evaluate portraits submitted by 6,000 contestants from 100 countries across the world, fueled controversy over racism. The contest was organized by Company Y, a deep learning research group sponsored by Company M, and primarily relied on AI algorithms which analyzes a database built with numerous portraits to judge beauty. It was known that objective indicators such as the symmetry of the human face and wrinkles were used as the criteria for judging.

What was the problem?

ase 2 C

The majority of the 44 final winners selected by AI were white females. Given that a significant number of participants were Beautiful. “colored people,” such a result was undoubtedly discrimination based on skin color. The technical director of the contest said, "If data on different skin colors is not sufficiently collected, in fact, AI can produce biased results." Although the technical staff did not set light skin as a requirement for beauty when designing the algorithm, the composition of the input data reveals that the Light AI was actually designed to reach such a conclusion. skin color 문제가 발생한 이유를 찾아 간략하게 적어 보자.

Is it possible to guess facial features just by hearing a voice?

Not beautiful.

Dark skin color

<Source> The Hankyoreh, Sep. 14, 2016

‘Speech2Face,’ a voice recognition AI technology which infers a person’s facial features based on intonation, pitch, words and pronunciation after hearing the voice, is attracting a great deal of attention. This technology was implemented by researchers from the Computer Science and Artificial Intelligence Laboratory at the Massachusetts Institute of Technology (MIT) who used millions of videos posted on a video platform operated by Company Y for training AI on the correlation between voice features and facial features.

What was the problem?

‘Speech2Face’ classified people with high voices as females and those with low voices as males. In addition, it identified the same Chinese male as Asian when speaking Mandarin and as Caucasian when speaking English. These cases exposed the technology’s bias against gender, race, etc. MIT explained that since Speech2Face was developed using the videos from Company Y, the distribution of the input data was not even.

Find the cause of the problem and briefly describe it.

<Source> AI Times, Apr. 28, 2022

Explore the filter bubble problem in the era of AI

‌Read the following passage about the movie <Ron’s Gone Wrong> and think about the society

with the problem of the filter bubble.

<Ron’s Gone Wrong> is a movie set in a world where high-tech social AI robots, ‘B-bots’, make friends with every child. This movie depicts the special adventure and friendship of

Barney, who receives a gift of ‘Ron,’ a defective B-bot that cannot connect to the network!

B-bot produced by the tech giant Bubble is a robot equipped with all functions needed to serve as children’s best friends, such as state-of-the-art intelligence, a user protection system, and social media streaming. Barney’s dad, who couldn’t afford to buy an expensive robot, made a secret deal with a delivery driver to buy a ‘B-bot’ that was damaged during the delivery process at a cheap price as a birthday present for Barney. Since ‘Ron’ is a broken product and cannot access the network, it does not follow the system’s algorithm. Thus, Barney teaches Ron ‘HOW TO BE MY FRIEND’ himself. This film shows the difference between friendships produced through big data and those built through communication. In this process, <Ron’s Gone Wrong> describes the duality of the information society with a sense of realism. The movie shows the positive aspects of the informatized society where users can enjoy the convenience from cutting- edge robots that protects them and freely share information while communicating with many different people. However, thorough this movie, we can also witness the adverse effects of informatization, such as people who share every moment of their private lives and find the value of their existence only in the number of ‘Likes’ and ‘Subscribes’ they receive from others, the possibility of privacy infringement due to the leakage of personal information, the spread of superficial human relationships with the decrease in face-to-face contact, and weakened consumer rights from information monopoly and control by Bubble. In particular, Bubble’s algorithm is designed to search for people with similarities in their daily life, food photos, wake-up habits, hobbies, etc., send a friend request, and recommend people who reply with ‘likes’ and ‘comments.’ In a world where friendship is made by selecting and recommending a friend based on data, instead of by sharing experiences together, it is difficult to have friends with different values or experiences. In the movie, Mark, the developer of B-bot, wants to make friends using this algorithm, while Andrew, who has become the CEO, aims to monopolize information and sell products by offering consumption experiences tailored to users’ search habits, purchase history, and location. Mark developed B-bot with the pure desire to help children make friends, but Andrew, as the company’s operator, only seeks profits, causing intense conflict between the two. This conflict shows that even AI technologies developed in good faith can be used for malicious purposes, and that we need to think critically when using AI technologies to protect our consumer rights.

Provide an operational definition of friendship

‌Give an operational definition of friendship to quantify its abstract concept and check out

different perspectives on ‘ friendship.’

Operational definition of a concept means the conversion of an abstract that is difficult to quantify into another concept that can be easily quantified. For example, the abstract

concept of ‘love’ can be defined as measurable concepts such as ‘number of dates in a week’ and ‘number of phone calls in a week.’

Expressing what “friendship” is in one sentence

List the properties that make up friendship

Provide an operational definition of the concept of friendship based on the properties listed in 2.

Properties

Affinity

Measure a friendship score

Properties

Affinity

You can understand the need to respect diversity in the use of AI technology.

You know that abstract concepts can be quantified through operational definition.

Operational definition

Define as the property of having a high number of meetings outside of school in a week, and measure each meeting as 1 point. e.g Give 1 point for 1 meeting and 2 points for two meetings with a total of 7 points

Friendship score

PMake a self-checklist for AI use

Draw up a self-checklist to practice respecting diversity for Bubble in the movie <Ron’s Gone Wrong>.

Jean-Philippe Vine, a co-director of the movie, said, “The world is totally saturated with technology. Technology has become a necessity for life, and people are increasingly exposed, through technology, to people and interests that are similar to their own.”

Write down why respecting diversity is required in the process of developing and using AI.

Why respecting diversity is required Momentum for social development, etc.

Among the 10 key requirements of AI ethics, select 3 requirements that you think are connected to respecting diversity and write down the reasons for your selection.

Requirement

The 10 key requirements of AI ethics are safeguarding for human rights, privacy protection, no violation, public nature, solidarity, data management, responsibility, safety, and transparency.

Reason for selection

Check out the top 3 core requirements related to respecting diversity selected by other classmates and why they were selected, and write down the core requirements that win the most votes from the classmates.

Number of votes

Number of votes

Number of votes

Safeguarding Human Rights

Privacy Protection

Number of votes

Reason for selection

Reason for selection

Public Nature

Solidarity

Number of votes

Reason for selection

Reason for selection

Responsibility

Safety

Number of votes

Reason for selection

Reason for selection

The three requirements our class selected as most relevant to respecting diversity are

No Violation

Number of votes

Reason for selection

Data Management

Number of votes

Reason for selection

Transparency

Number of votes

Reason for selection

4 Make an AI self-checklist for Bubble based on the selected core requirements.

Diversity

Transparency

Does this technology fully reflect human diversity and ensure that all people are treated fairly?

Was a specific group excluded in the training of this technology?

5 Make an AI self-checklist for users based on the selected core requirements.

Safety

You can select requirements relevant to respecting diversity and write down the grounds for your selection.

Are there any procedures to perform consistent assessment on the safety of the results from

AI systems?

You can propose a self-checklist for the use of AI technology.

UNESCO “Recommendation on the Ethics of Artificial Intelligence”

(November 2021)

<Preamble> The General Conference of the United Nations Educational, Scientific and Cultural Organization (UNESCO)

meeting in Paris from November 9-24, 2021, at its 41st session, recognized the profound and dynamic positive and negative impacts of artificial intelligence (AI) on societies, environment, ecosystems and human lives, including the human mind, in part because of the new ways in which its use influences human thinking, interaction and decision-making and affects education, human, social and natural It noted that addressing risks and ethical concerns should not hamper innovation and development but rather sciences, culture, and communication and information, provide new opportunities and stimulate ethically-conducted research and innovation that anchor AI technologies in human rights and fundamental freedoms, values and principles, and moral and ethical reflection, recommends that Member States bring the Recommendation to the attention of the authorities, bodies, research and academic organizations, institutions and organizations in public, private and civil society sectors involved in AI technologies. 19. <Ensuring diversity and inclusiveness> Respect, protection and promotion of diversity and inclusiveness should be ensured throughout be done by promoting active participation of all individuals or groups regardless of race, color, the life cycle of AI systems, consistent with international law, including human rights law. This may descent, gender, age, language, religion, political opinion, national origin, ethnic origin, social origin, economic or social condition of birth, or disability and any other grounds. 20. optional use of AI systems and the co-design of these architectures should not be restricted during The scope of lifestyle choices, beliefs, opinions, expressions or personal experiences, including the Furthermore, efforts including international cooperation should be made to overcome, and never any phase of the life cycle of AI systems. 21. take advantage of, the lack of necessary technological infrastructure, education and skills, as well

landlocked developing countries and small island developing States, affecting communities. as legal frameworks, particularly in low- and middle-income countries, least developed countries

<Source> “Recommendation on the ethics of artificial intelligence:” Korean National Commission for UNESCO. Artificial Intelligence (AI) Ethics and Law - (I) Study on issues of AI ethics and governance

Different from each other, not the same

Treating people unfairly for “being different”

Tendency to favor one side over another

A culture that is enjoyed by a majority of the members of society

2. Take initial letter quizzes and review key terms.

Discrimination

Difference

Mainstream culture

Bias

‌ ‌ 1 If the AI system imitates and reproduces the stereotypes and prejudices accumulated in our society and culture, the development of AI can eventually become the epicenter of new social discrimination and trigger further social conflicts. Therefore, we need to respect [D ].

‌ [F ] [B ] is a phenomenon of ill-balanced acceptance of information caused by recommendation 2

algorithms and implies that users are confined to only the information category filtered to suit their interests.

‌ 3 [C ] [B ] refers to a cognitive bias that seeks and accepts only information or evidence that are in line with one’s belief and ignores other information that contradicts the belief.

You can understand the concept and importance of respecting diversity in the context of using AI.

Protecting Privacy

When biased results are produced from AI-based decision-making, you can assume this problem in connection with data bias.

You can develop and suggest a self-checklist to respect diversity in the deployment of AI technologies and services.

Ⅳ

Doing No Harm

학

You will be able to look at various violation cases related to AI systems and find solutions for them. You will be able to understand that developers, users, and policy makers in the event of any violation due to the use of AI systems, need to make shared efforts to compensate for damages and prevent any recurrence.

maker Policy

You will learn that AI should not be used to harm humans, and discover ways to deal with the risks and negative consequences that can be triggered by AI.

Developer

User

‌Although AI is developed for humans, it sometimes harms humans, defeating the purpose. What efforts should be made to prevent these issues?

Oh, my god! Is it really them?

Yes, I am so happy to see them again.

Really? Were they your favorite band? Wow, that is

Well. They would think they are working even in the afterlife. Also, they are not being paid any.

What?

I can’t believe I am able to watch my favorite singers on the stage again. I was a huge fan of theirs before they

passed away.

believe this is Mom, do you what they wanted as well? The deceased?

interesting!

Hmm, that makes sense. Who will take all the money earned from their performance?

Types of violations from AI

Misuse of data

Stop it! You should observe the regulations.

What is ‘no violation’ in the context of AI?

No violation means that we should pay more heed to the prevention of any infringement on the safety of our society and the rights of individuals in developing and using AI, and to predict and prevent infringements in advance. AI system is a technology developed for the benefits of humans. Therefore, AI systems should be designed to prevent the infringement of basic human right is infringed through prior investigation of various types of potential damage, including physical and mental harm, in the development phase. Even if AI systems are designed with possible infringement of human rights considered, they may cause damage to users if they are used beyond their intended purpose or if direct or indirect risks occur due to unintended or unexpected factors. Hence, we should seek solutions to address damages from AI and minimize the potential risk factors of AI system by looking at various cases of damage caused by AI systems.

Violation of basic human rights

A consulting company in the US has categorized five types of violations that may take place due to errors or misuse of AI systems.

This is a type of infringement caused by misuse of 2 1 data for AI, violation or misinterpretation of security

regulations, and occurs when legal provisions are violated in collecting and using data.

Use only for intended

purposes

Examine any potential damage to users

What else should be taken into account?

Failure of management

This is a type of violations that occurs in relation to various negligence and improper management that can be caused by the AI system manager in using AI systems.

I am sure AI finished all the work

Flaws in AI

envi ro nmen ts andd ata processing, and can result in prolonged AI service failures. It may also occur when services are provided based on erroneous data or severely biased data.

what??

What happens in the process of collecting and processing big data

Company

Video sharing platform company Y

Attack by malicious programs

This type of violation occurs 4 malicious programs designed for 3 in inadequate AI training information leaks, data extortion, work disturbance, and other criminal purposes. In this case, a violation is caused by viruses, malware, ransomware*, etc. that can infect or modify AI systems and machine learning. Let’s make the AI system flooded with errors!

Based on the data, A black person is false!

Big data is created through a collection of traces left from the use of the internet or social media. For example, the online shopping mall of a fashion brand Z records data about every visitor’s products of interest, preferred colors and patterns, etc, regardless of his/her actual purchase. Even in the offline stores, the brand puts RFID* tags on every piece of clothing, collecting data such as what clothes customers wear most in the dressing room, what clothes they sell the most, and so on, and summarizing all of this data into the next design.

Major numerical data generated in a minute The world ’ s largest online shopping mall A A customer spends $283,000 on Company A. Free message service platform company I 12 million people send and receive messages. 694,000 videos are streamed. Short-form video platform company T 167 million videos are viewed.

Online photo sharing and social network service company I 65,000 photos are shared. <Source> Visual Capitalist Amount of data created on the internet in a minute as of 2021

Thus, big data can help companies strengthen their competitiveness. In addition, if the government identifies citizens’ needs through big data and establishes and implements customized policies, social costs will be reduced and the quality of public services will be improved.

Abuse by the government

AI systems may be attacked by 5 ethical standards are not complied

It refers to cases where legal and Doing No Harm policies using AI. For example, with in the process of enforcing machine learning and AI technology could be used to enhance a government’s monitoring of its citizens, which might lead to the problem of surveillance society*.

* Ransomware It is a compound word of ‘Ransom’ and ‘Software’ and refers to a malicious program that demands money after locking the system or encrypting data to make it unusable. * Surveillance society A surveillance society is a society that collects, analyzes

and uses information about its individuals or groups extensively like “Big Brother” in George Orwell’s novel 1984.

* RFID It is a device that stores product information in a very small chip using radio frequency and transmits data wirelessly.

Why is it looking at me?

However, there are also concerns that the use of big data could have the side effect of giving birth to Big Brother* which monitors and controls society by monopolizing all information, as described in George Orwell’s novel 1984. A person in a documentary film <Citizenfour> stated, “The Americans’ justification for everything since the September 11 attacks is terrorism. Everything is in the name of national security, to protect our population. In reality, it’s the opposite. A lot of the documents have nothing to do with terrorism or national security, but with competition between countries and with companies’ industrial, financial or economic issues.” He said that although the US government is promoting the war on terrorism as a justification, data is actually used for the benefit of the country and corporations. He pointed out that there is a risk that personal information and privacy will be infringed and misused in the course of collecting and processing big data.

It seems that he knows everything about what I listen to.

I feel like I am being monitored. It is scary.

Examples of infringement of personal information and privacy in the process of collecting and processing big data

Actor-specific efforts required to prevent infringement

Maintaining AI systems as a technology that guarantees users’ fundamental rights requires relevant economic actors: companies, individuals, and governments to predict possible ethical issues and to pay consistent attention to them from the development stage to the utilization stage. Ethical consciousness of AI is a prerequisite for companies and developers making AI. Moreover, the ethical judgment and attitude of individuals using AI services also play a significant role in developing and improving AI for the benefit of humanity. For governments, it is necessary to prepare legal provisions and systems that can relieve damage caused by AI that may occur in the private economy sector, and to seek ways to strengthen citizens’ ethical consciousness of AI. 068

* Big Brother Big Brother refers to the absolute power in George Orwell’snovel1984,a watchdog who controls society by monopolizing all information. In 1984, Big Brother’s official purpose to maintain social order, but it is actually used as a means to maintain power.

Deepfake videos used in war

‌Review the key concept you learned earlier with relevant cases.

In wartime, some hackers create and distribute deepfake videos that are precisely manipulated using AI technology to amplify the uncertainty, confusion and distrust of the enemy. How should we respond when falsified messages about a politician surrendering to war spread on social media?

How should we respond to problems after their occurrence?

Relevant social media companies immediately deleted videos and posts containing falsified news, and the government quickly announced that the information was not true to stop spreading confusion. Citizens need to develop their critical thinking skills and media literacy skills to accept information and knowledge in a critical manner.

<Source> Herald Business, Mar. 17, 2022

Write down the infringement issue in this case and the countermeasures required for each pary involved.

침해 금지

Understand the interactions between user and developers through Activity AI-powered technologies or services.

‌Based on data on AI technology, find AI technologies that are used in real life. Also, think about what you are provided with and what you provide through the use of AI technology or service, respectively, and analyze the benefits and anticipated harms of such mutual exchange.

Both interest andconfi-

60s

dence in AI were found to

50s

be high in the elderly group.

20s

These are the results of a survey of 3,500 people aged between 14 and 65!

AI translator

●

AI-enabled car navigator

●

Chatbot

●

AI personal assistant

●

AI speaker

●

Hospital/ Medical Treatment/ Health Care 62.1%

Current status of AI popularization

Level of interest

60s

65.9%

10s

62.9%

50s

60.0%

Top 3 age groups with the highest interest and confidence in AI

46.0%

35.5%

33.4%

Top 5 AI products and services with the highest usage rate

Disaster/ Quarantine

Transportation/ Distribution

33.0%

27.9%

Top 5 fields with the most urgent need for AI introduction (multiple responses)

Level of confidence

57.6%

43.8%

43.5%

Currently in use

Once used but not anymore

55.7%

23.5%

55.8%

17.7%

26.9%

36.1%

26.1%

Public Safety

Manufacture

27.0%

26.4%

<Source> The Fourth Industrial Revolution Committee, Survey on the use and perception of AI for its popularization, Jun. 2021

AI product or service that you use most frequently

Similar song recommendation on a music streaming service

What are improvements you think necessary to the AI products you use?

Think about what you are provided with and what you provide through the use of AI

technology or service, respectively, and analyze the benefits and anticipated harms of such mutual exchange.

AI translator

AI-enabled navigator

AI personal assistant

AI speaker

Others

You can understand the limitations and problems of AI technologies or services that are widely used in real life.

You can identify and analyze data exchanged between users and developers in the use of AI technologies or services.

What you are What you provide provided with

Translation of ‌C orpusdata,myvoiceand ‌A Itranslator that recognizes ‌Conversation is recorded while foreign languages colloquial corpus such as newly dialects, easing oflanguage the microphone is on to use the

coined words

Details

You can predict future positive effects and infringement situations resulting from the data collected in the process of using AI technology or services.

Benefits from mutual exchange

Anticipated infringement

barriers, etc.

translation service.

Analyze the case related to face recognition technology ‌Analyze the case of an incidents caused by errors in the face recognition system, and think about the rights violated and the proper remedies.

In January 2020, Mr. A, a black man living in Detroit, was spotted as a suspect of a watch theft from a luxury store. In this case, he was arrested by the police in the front yard of the

house with his wife and daughters looking the scene, but it later turned out that he was innocent, but he had been arrested due to an error in the Detroit Police Department’s face

recognition system.

Think about the case from the respective perspectives of Mr. A and the police, the parties concerned, and write your opinion.

Case Overview

No, I am innocent. I hope you don’t think all black people

You are a In March 2019, the Detroit Police Department put criminal. You are under a fuzzy photo of a black man in a red hat from arrest. CCTV into a face recognition system developed by Company D to contrast it with photos from driver’s licenses stored in national photo repositories across the United States. As a result, it was discovered by the system that the photo of the suspect and the photo of Mr. A’s driver’s license matched. Afterwards, the investigators added Mr. A to the list of suspects, and the watch store’s security officer, who was not present at the time of the theft, identified Mr. A as the culprit of the theft. Detectives issued a warrant that required the signature of a high-ranking executive of the police department and arrested Mr. A. After checking the photo of a suspect presented by a detective of the Detroit Police Department during interrogation, Mr. A said, “This is not me. I hope you don’t think all black people look alike.” However, the detective replied “The computer says it’s you.” Consequently, Mr. A had to spend 30 hours in a detention center.

If I were Mr. A?

look alike.

If I were a police officer?

Analyze the points in dispute of the case and the positions of both sides, and think about the

Main Issues of the Case

violation of rights and applicable remedies under Korean legal provisions

The Detroit Police Department relied on the biased and flawed face recognition technology to arrest Mr. A. Mr. A’s basic human rights guaranteed equally to everyone were violated due to his race or skin color, and flaws in the use of AI technology were revealed.

A’s position on the lawsuit

Mr. A filed a lawsuit with the support of the American Civil Liberties Union (ACLU). “Detroit police have identified the culprit without sufficient investigation of other evidence or information, only relying on flawed and racist face recognition technology,” Argued ACLU. “Governments’ use of face recognition plainly has a disparate impact against people of color, Study after study has shown that this dangerous technology has far higher rates of false positives for people of color compared to white people. Thus, government use of this technology violates laws that prohibit government from adopting practices that cause disparate impact.” Through this lawsuit, Mr. A seeks compensation for damages, higher transparency on the use of Detroit Police Department, whether directly or indirectly. face recognition technology, and a complete discontinuation of the use of the technology by the

Detroit’s position on the lawsuit The counsel for the City of Detroit said “As the police chief has explained, the arrest was the result of shoddy investigation - not faulty technology. The Detroit Police Department has conducted an internal investigation and has sustained misconduct charges relative to several members of the department. New protocols have been put in place by the Department to prevent similar issues from occurring.” He argued that the city aimed to achieve resolution in the case, but face recognition was not to blame for the situation.

Statutory provisions on infringement of rights and remedies in Korea

Refer to the provisions on infringe- ment of rights and remedy, and discuss what decision should be made in this case through group discussion.

person who causes losses to or inflicts injuries on another person by an unlawful act, intentionally Article 850 of the Civil Act of Korea regulates the liability for damages by stipulating that “any

First

Second

Third

In order for damage from a tort is constituted, the following requirements should be fulfilled. or negligently, shall be bound to make compensation for damages arising therefrom.”

intent or by negligence. The act of the perpetrator should have been committed with It refers to a special reason that

perpetrator commits an illegal act, he/she may not be held legally The act of the perpetrator should be illegal. However, even if the

responsible for the illegal act if there is a justifiable reason*.

even if the requirements for a criminal act or tort are formally Damage should occur and there must be a causal relationship between the perpetrator’s act and such damage. In this context, fulfilled. This includes cases in

damage includes not only property damage but also mental damage. which damage is unavoidably

* Justifiable reason

causes the relevant act to be deemed not substantially illegal

Fourth, It should be acknowledged that the perpetrator has the capacity for assuming responsibility. If the perpetrator’s capacity for inflicted on others, such as self- defense against another’s illegal

assuming responsibility is not acknowledged, as in the case of a child or a mentally ill person who lacks the judgment ability, he/ acts, emergency evacuation

she shall not be held accountable for damages.

In principle, compensation for damage should put the victim back to its original state before the damage occurred. However, in reality, it is difficult to restore the original state, damage and mental damage. In the case of a tort which has defamed so the main method is to make monetary compensations both for property one’s reputation, the court may order the perpetrator to take measures appropriate for repairing the injured party’s reputation, together with provision of compensation. However, the publication of apology in newspapers is not allowed.

to avoid an imminent crisis or disaster, etc.

Write a scenario for a mock trial

‌Refer to the following responsive measures taken by various sectors in the US against the face recognition technology for a mock trial scenarios, and after having discussions in groups about

the appropriate judgment for this case, write a mock trial scenario.

Responses of each field in the United States to face recognition technology after the incident

(excluding federal airports and ports).

City of San Francisco Government • In May 2019, as a first attempt in the Local United States, 53 city departments, including police, were banned from using face recognition technology

development or sale of face recognition technology).

City of Boston • In June 2020, a bill banning the use of face recognition technology by under the City was passed. any and all entities and practitioners

State of Washington • In March 2020, a legislation to regulate face recognition technology was enacted for the first time among the US federal and state governments. • The use of face recognition technology for continuous surveillance, real-time or quasi-real-

agencies is banned (not applicable to commercial time identification, or continuous tracking by state

City of Portland • In September 2020, a bill banning the use of devices with face recognition technology in all businesses and public institutions was passed (taking effect on January 1, 2021).

• In February 2020, a bill for the ‘Ethical Use of Facial Recognition Act’ was introduced. Based on criticisms that face recognition technology is less accurate for minorities, such as women and people Assembly Federal of color, the bill prohibits federal agencies from using face recognition technology until the guidelines are established. • In June 2020, a bill for the ‘Facial Recognition and Biometric Technology Moratorium Act’ was introduced. The acquisition, possession, access, and use of biometric surveillance technology by

recognition and gait recognition are included). federal law enforcement agencies is prohibited as illegal (In addition to face recognition, voice

• Politician has repeatedly called for addressing algorithmic biases of face recognition technology based on race, gender, or other characteristics. Political • Politician propose to enact a biometric data protection law at the federal level, stressing the need to Arena regulate face recognition.

association Academic • In June 2020, ACM urged the US Congress to ban the use of face recognition technology by

Company I Assembly Federal • In June 2020, the CEO of Company I pointed out that face recognition technology could be misused as a discriminatory tool, and announced the company’s exit from the face recognition business.

Company M

Company T •I nSeptember2020, Company T expressed • In July 2020, Company

• In June 2020, Company M clearly stated that it would not deliver the service to police until federal regulations on face recognition technology is prepared (prior to the announcement, the company was supplying its face recognition service API to the police).

its apolog y to users who criticized that the image algorithm instigated company’s automated racism due to its preference for photos of whites over blacks (when face photos of a black and a white • With the goal of making person are uploaded simul ta neously,the company’s thumbnail algorithm preferentially selects and presents the white person’s photos).

Association for Computing Machinery (ACM)

Company A

• In June 2020, the company said it would suspend its face recognition service product introduced by the police for a year.

Company M M launched a dedicated team to address racial discrimination issues within its services, such as improving algorithms that encourage racial discrimination.

social media fair and safe for all users, regardless of skin color, the company improved the racial bias in its algorithms used for its social media.

governments and corporation. (However, this is not a request for a permanent ban, but a request for a temporary suspension until an accurate standard is established.)

American Civil Liberties Union (ACLU) • The ACLU has been calling for a halt to the use of face recognition technology for years, citing the risk of bias and privacy violation, and has filed lawsuits against numerous companies. (The federation shared the results of a 2018 experiment in which a face recognition program identified 28 (mostly people of color) out of 535 lawmakers as criminals, criticizing the imperfections of the technology.)

group Civic

<Source> Social Controversy and Implications of the Introduction of Face Recognition Technology: Focusing on the US Cases (ETRI Technical Policy Brief, Dec. 2020)

‌Based on 1

Plaintiff

Defendant

Judge

and responsive measures taken by various sectors in the US against the face recognition technology after the incident, write a mock trial scenario and conduct the mock trial.

Let’s list arguments the plaintiff might give!

Let’s list arguments the plaintiff might give!

Let’s list arguments the plaintiff might give!

Assign the roles of script writer and characters for the mock trial (plaintiff, defendant, plaintiff’s counsel, defendant’s witness, presiding judge), and start the trial!

Write a scenario in light of various elements including plaintiff’s or defendant’s argument, summary

of issues, application for evidence, examination of witnesses, final

argument, and judgment!

(

100%?

Write a statement of opinion of a student participant to the mock trial.

Statement of opinion of a student participant to the mock trial

), a student participating in this mock trial, hereby states my opinions on this case as follows.

The defendant (is / isn’t) liable for damages on the grounds that ( )

With respect to the plaintiff’s claims for damages, greater transparency on the use of face recognition technology, and the Detroit Police Department’s cessation of the use of face recognition technology, I think that ( ).

1. If you were a judge, how would you determine the percentages of liability between the plaintiff and the defendant, given that the total liability rate is

- Plaintiff’s percentage of liability ( liability ( %)

2. If you were the judge, do you think that the defendant is liable to pay damages to the plaintiff? ( Tick the relevant check box ▢.) - ▢ Liable ▢ Not liable

You can analyze damage cases related to AI from various perspectives.

You can suggest measures to compensate for damage and prevent recurrence in the event of any damage from the use of AI systems.

%): Defendant’s percentage of

‌ ‌ 1 [S ] [S ] is a society that collects, analyzes and uses information of its individuals or groups extensively.

‌ 2 [R ] is a malicious program that demands money after locking the system or encrypting data to make it

unusable.

2. Think of and write down five types of violations that can arise from errors and misuse of AI systems.

3. List the advantages and disadvantages of using big data.

No Violation

You can explain the concept and meaning of no violation with respect to the use of AI.

You can propose technical and social institutional measures against infringement situations resulting from AI technology.

You can analyze the infringement situation caused by AI technology, determine who should be held responsible and find solutions to the problem.

Ⅴ

Working for the Common Good

You will be able to understand that the development and use of AI can make contributions to the greater happiness of individuals and strengthened public natures of the society. You will be able to understand that basic human rights may be infringed by the development and use of AI. You will be able to explore cases where the development and use of AI has led to the greater happiness of individuals and strengthened public natures of the society and where it has failed.

Let’s have a conversation with AI translator for

sign languages!

You will understand that the development and use of AI should contribute not only to the pursuit of individual happiness but also to the promotion of the society’s public nature, and discuss ways to maintain the public nature through case studies.

‌Although there is an overall consensus at home and abroad that the development of AI technology should be in harmony with the improvement of the safety and welfare of humanity as a whole, we witness cases in which the public nature is damaged in the development and use of AI. What measures can be taken to promote the public nature?

My dad’s car fell into a pothole on his way to work, so he almost got into trouble.

I gave some thought to the accident, and realized it would be great if we could use AI technology to monitor the road from time to time

and inform us of risky things like potholes in real time.

Oh, just like a smart city? Don’t you know? It is a city that uses electronic sensors for data

collection to provide the necessary information public good, but I think What? for efficiently manage Smart city? assets and resources.

Really? It could have been a disaster.

Wouldn’t it be nice to have a smart city? People will be safely protected and enjoy

convenience. It must provide advantages for the

it also has drawbacks, doesn’t it?

AI and public nature*

Please do your best for me!

As implied in the Thomas Kuhn’s remark that “Today, the speed of technological development is phenomenal, and the effects of technology is transforming our lives,” AI, which once only imitated human intelligence, is now outperforming it in many areas thanks to the advent of deep learning and further advancement of technology. Deeply embedded in our daily lives through intelligent products and services, AI is addressing serious social challenges and expanding its domain. In addition, AI is already being deployed in smart cities and other fields for the convenience of members of society. Numerous advanced AI technologies, such as Watson, AI for medical use developed by Company I, large language model GPT-3 produced by an AI research business, AI for learning, and AI for online unmanned stores from Company A, are promoting innovations in all areas of society, beyond the technological field.

You are Your diagnosis suffering ○○. is quick and accurate.

Sure!

Use of AI technology in the medical field

Each customer is recognized

The payment is made

creates text by The algorithm

automatically.

automatically after shopping.

connecting words and phrases based on numerous texts from the internet learned. It is very convenient to use!

AI for learning and AI in an offline unmanned grocery store

* Public Nature It refers to a nature that is not limited to one individual or group but is comprehensively related to the entire members of the society.

Efforts required to ensure the public nature is considered in the use of AI

AI systems should not be used only for the benefit of limited individuals or groups. Rather, it should be used for the interests of society as a whole. As members of the AI era, we should seek ways to develop AI that can strike a balance between individual happiness and the public nature of society.

Development of AI technology and ethical vacuum

As AI technology develops rapidly, we are facing ethical issues and dilemmas* that did not exist before. When the pace of development of ethical regulations fails to keep up with that of technological development, ethical vacuum may occur. In the process of designing and developing AI technology, ethical issues that undermine the public nature for the sake of profit creation may arise, and decisions made by AI systems may have discriminatory or unfair influences on certain classes. Therefore, we need to achieve a social consensus that efforts to secure the public nature in the development and use of AI systems are required, and prepare appropriate standards and supplementary measures.

AI systems have a great impact on our lives at the center of rapidly evolving technological development. The influence of AI is not limited to a specific target, and it is humans who should bear the responsibility for the results caused by the AI system. How can we ensure that AI is used to the greatest benefit of the overall humanity? Users, developers and policy makers should make collective efforts to prevent ethical issues arising in the process of developing and using AI technology from being ignored and to promote the public nature of society.

I support the development of crosswalks for the visually impaired!

AI User

Efforts required to maintain the public nature of AI systems

* Ethical Dilemma It refers to a situation in which the value to be prioritized should be determined among various duties and missions in the form of specific behaviors.

Let’s develop crosswalks for the visually impaired to serve the public interests!

AI Developer

Let’s devise plans to increase public interests through international national and

conferences!

Maker Policy

Voice-enabled traffic lights with IOT technology

◯◯ City announced its plan to add ‘smart sound signals for the visually impaired’ equipped with Internet of Things (IOT) technology to traffic lights for pedestrians at crosswalks in the downtown. A ‘smart sound signal for the visually impaired’ is an apparatus that delivers the walking signals through voice to help pedestrians with disabilities, including the visually impaired, can safely use the crosswalk. The smart sound signal for the visually impaired is furnished with IOT technology and Low Energy Bluetooth function, so users can operate it using a remote control for the visually impaired and a smartphone application without operating it in person. In addition, through real-time remote monitoring, it allows quick inspection and repair of malfunctions in power, buttons, speakers, etc., which could only be verified by on-site inspections or civil complaints in the past. With the aim of achieving a 50% supply rate of sound signals by 2025, ◯◯ City plans to continuously carry out the sound signal installation projects for the improved walking convenience and prevention of safety accidents for the vulnerable, including the visually impaired.

What Could Be the Problem?

<Source> Anjun Journal, Apr. 11, 2022

Korea has adopted a manual sound signal which generates a signal only when a button on a traffic light is pushed in person. In this method, the visually impaired have to find and press a button to hear the sound, which causes inconvenience to them as pedestrians and serves as a risk factor. In addition, since no standard for installation has been put in place, such as the appropriate location of sound signals, we often witness signal speakers installed on flower beds away from crossings, or signal buttons installed far away from braille blocks.

Where on earth is the sound signal button?

<Source> The Kookje Daily News, Apr. 19, 2021

What’s wrong? I thought I have correctly followed along the braille

blocks.

With respect to AI-enabled products and systems, look for and summarize cases where the public nature can

be enhanced.

What if we could communicate with Deaf people without knowing the sign language?

Sign language is a visual language that communicates thoughts using physical signals, including hand movements, instead of voice, and also refers to a linguistic system that uses a sign language. Sign language is formed by combining the shapes drawn by fingers or arms, their position or direction of their movement, facial expression or movement of lips. Sign language interpreters are often involved in communicating with hearing-impaired people. Sign language is generally used by people with hearing impairments, and non-disabled people often have little experience in learning or seeing sign language in real life. However, some non-disabled people learn and use sign language to communicate with hearing-impaired people depending on necessity or interest. How about normal circumstances? If you have to use sign language to communicate with the friend sitting next to you at school, it will be quite inconvenient. Students at ◯◯ High School who were looking for a way to reduce this inconvenience created a speaker that uses AI technology to translate sign language into normal language. The students made a data image by filming sign language movements in person with colored gloves on. They developed a program that recognizes colors in data images and creates afterimages, and collected data on the coordinates (x, y) of moving hands. After several students repeatedly filmed the same movements of sign language, they collected data in the same manner. As a result, an AI speaker using an algorithm that efficiently obtains data was completed. Later, the students improved the algorithm model to mathematically calculate the coordinates of the joints, and then developed a new AI translator.

Why Do We Need AI for Translating a Sign Language?

Check out the development process of the AI sign language translation from the developers at ◯◯ High School at the link below!

https://www.youtube.com/watch?v=dm_mj20HAO8

In early 2021, ◯◯ City began offering the world’s first civil complaint guidance system service with AI-based sign language motion recognition technology. ◯◯ City officials explained the background of the system development as follows. “Existing automatic guidance systems in public facilities were text-based, making it difficult for visually and hearing impaired people to use. In particular, it is often very difficult for hearing-impaired people who use sign language as their daily language to understand text.”

<Source> Gonggam, Jan. 24, 2022

Think about why AI-based products or services such as sign language translation speakers using AI are

developed, and briefly write it down.

공공성

Explore the public nature of AI systems and investigate related cases.

‌Read the following articles about satellite image analysis and investigate cases online where AI technology is

used the common good of society and humanity.

Deployment of machine learning technology to analyze the size of the economy in impoverished areas

Sciences at Stanford University, USA, published in the international A team of Professor Marshall Burke of the Department of Earth System economy in impoverished areas by combining artificial satellite imagery journal ‘Science’ the technology to accurately analyze the size of the and machine learning technology, which was developed in partnership with the Center for Food Security and Environment at the same university and the National Institute of Economic Research (NBER) Accurate identification of household income and expenditure by developing countries to determine the exact size of their economies, as region is crucial in policy making. However, it has been difficult for Due to this reason, in the past, a method of analyzing night time photos necessary investigations are costly and time consuming. at the intensity and width of the light in the night map image observed was used. This method estimates the size of the economy by looking only international poverty line.

‌Areas marked in dark red indicate the regions of extreme poverty below the However, this method was only able to find out a rough trend, but it was not enough to infer detailed information. by satellite under the assumption that the more lights are on at night, the more active the economic activity is. Particularly in poor countries, there is little light at night, making it difficult to analyze regional differences. To overcome these limitations, the researchers used AI to analyze satellite images during the day. This is and relatively large amounts of information can be collected. because, during the daytime, it is easier to capture the movement of vehicles, hunting and farming activities, First, the researchers fed satellite images of several countries, such as Europe and Asia, where the size of the economy is already known, together with the data on their sizes of the economy. The AI, which was trained on The researchers leveraged Convolutional Neural Network (CNN) technology, which is mainly used for image this data, estimated the size of the economy by itself when a satellite image of a developing country was input. distinguished between paved and unpaved roads, metal roofs and thatched roofs, and other element reflecting identification, as well. The technology was used to break down and classify AI images into 1,000 categories, and In addition to this, the research team provided AI with artificial satellite images of Company G’s map, which the characteristics of impoverished areas. captures the footages of the entire earth, as training data. As a result, the AI found areas below the poverty line, which represents the lowest standard of living, 81% more accurately than when using only nighttime satellite images. In addition, the AI was 99% better at identifying areas that were less than half the poverty line. The researchers used this technique to estimate the size of the economies of five countries: Nigeria, Tanzania, Uganda, Malawi and Rwanda. The analysis found that most of Malawi’s land area had an economy below the poverty line. It was also verified that the living standards of people in Tanzania, Uganda and Rwanda were near Professor Burke said, “We have verified that AI can accurately identify up to 75% of changes in economic the poverty line. activity in a specific region. This finding is of great significance in that it helps us accurately understand the economic conditions of poor countries that have been hidden in a veil of mystery.”

<Source> Dongascience, Aug. 2016

Investigate case of promoting individual happiness or the public nature of society. Look on the internet for examples where AI is used and discuss in the boxes below how this

Individuals’ Happiness

Increased Public Nature of the Society

technology promotes individual happiness or the good of society as a whole.

Positive Effect

Negative Effect

Examine the social and economic impacts

Write down the efforts needed to minimize the misuse and abuse of AI technology

Choose a topic

Organization of groups and assignment of roles 2

Debate for securing the public nature of AI

‌Based on the cases related to the public nature collected in

each actor to maximize the positive functions of AI and minimize the adverse ones.

Select one of the cases you investigated in to public natures of AI. ‌

developer, and a policy maker).

, discuss the efforts required for

Activity 1

in relation

Activity 1

Organize a group of 3-4 people, and assign actor roles within each group (Select 1 among the roles of an AI user, an AI

Design of a discussion strategy and preparation of relevant materials Do research on each selected actor, design discussion strategies through group discussions, and prepare related materials. 3

Initiation of discussions and preparation of questions 4

provision of feedback Wrap-up of the activity and 5

Choice of topic

Discussed Case

Group Composition

Actor Role

related to the discussion. students listen to the discussion as a panel and write questions

provide feedback on the overall discussion activity.

Organization of groups and assignment of roles

• Name of the group

• Role of participants

Conduct group discussion activities from the perspectives of an AI user, an AI developer, and a policy maker, and the rest of the

positive functions of AI and minimize the adverse ones, and Summarize the efforts required for each actor to maximize the

Discussion Materials

• Role

• Relevant questions

Design of a discussion strategy and preparation of relevant materials

Initiation of discussions and preparation of questions

Participant Materials for Discussion

• Claims and supporting facts

AI User

Wrap-up of the activity and provision of feedback

AI Developer

Prepare discussion materials focusing on the efforts that Working for the Common Good each of an AI user, an AI developer, and a policy maker should make to maintain public nature!

Policy Maker

Summary of final plans to maximize the positive functions of AI and minimize the adverse ones

You can find cases where AI technology is used for the common good of society and humanity.

You can propose ways to prevent the damage to the public nature and minimize the misuse or abuse of AI system when using it.

Carry out a project to develop an AI with the consideration for the public nature.

‌Based on the design thinking strategy, design an AI program by following the process of developing an electric wheelchair for a quadriplegic that can be controlled with the face recognition function.

Design Thinking is a creative strategy that designers utilize in the design process. It is an approach that can be used to consider problems, to solve problems more broadly, and it has been applied to address industrial and social problems. Design Thinking uses the designer’s sense and method of turning a technically available, viable business strategy

Empathize

into customer value and market opportunities to meet people’s needs!

3. Ideate

1. Empathize

2. Define

5. Test

4. Prototype

Search and choose a topic or problem that you empathize with. The Problem You Empathized with (Example) According to the National Spinal Cord Injury Statistical Center, it is estimated that approximately 288,000 people in the United States suffer from a spinal cord injury, and there are nearly 17,700 new cases each year. A study conducted in 2018 verified that physical mobility has the greatest impact on quality of life for people suffering spinal injuries.

Define

Define the details of the problem discovered in the Empathize step and set a goal. Problem to Define (Example) Quadriplegic patients are usually only able to move around with the help of a caregiver or by relying on an electric wheelchair that has complex sensors attached to the body and whose operation method should be learned separately. How can you help people with limited mobility regain full mobility and control? Is there any way for them to move smoothly without any difficulty?

Present ideas to solve the problem you’ve Working for the Common Good defined, using various methods, such as brainstorming.

Come up with ideas Quadriplegic patients can select and use 10 different facial expressions to control the electric wheelchair including moving forward, turning and stopping.

Prototype

You can describe your prototype in writing, express it as a picture, or use a storyboard!

In a short period of time, describe your abstract ideas concretely and visually, regardless of format.

Program design •‌ Create models by training various facial expressions and then make a prototype using an app that can be tested.

patients can drive themselves.

This electric wheelchair can be controlled

This is a wheelchair that quadriplegic

using 10 different facial expressions of the patients!

Computer vision is a field of AI that uses computers to reproduce general visual perception abilities of humans.

Present

• Is there any risk of a serious accident if a quadriplegic patient misunderstands or forgets the facial expression for the direction he or she set? • Wouldn’t there be any problem in creating technology using only computer vision?

Modify and supplement the shaped prototype.

Share and present the details of the project using Jamboard. Using Jamboard •‌ You can make presentation materials for the design thinking application project while sharing them with your group members on Jamboard.

Decorating Jamboard with details

Sharing a link using the editor function Giving feedback using the sticker note feature

Empathize

Define

Problem to Define

‌Refer to the example and design an AI system to promote the public nature.

The Problem You Empathized with

Come up with Ideas

Prototype

Program Design

Present

You can design and plan AI technology to promote the public nature through design thinking.

Decorating Jamboard with details

Sharing a Link Using the Editor Function

Our efforts to secure the public nature of AI

You can explore areas where AI can be used to promote the public nature, and explain the efforts required to secure the public nature of AI.

Giving Feedback Using the Sticker Note Feature

‌ ‌ 1 [E ] [V ] is caused when the pace of development of ethical regulations does not keep up with that of

technological development.

‌ ‌‌ [E ] [D ] refers to a situation in which the value to be prioritized should be determined among various 2

duties and missions

‌ ‌ 3 To ensure that AI is used to the fullest benefit to humanity as a whole, ethical issues arising in the process of

developing and using AI technology should not be disregarded, and [U ], [D ] and [P ] [M

] should make collective efforts to promote the public nature of society.

2. Summarize why public natures should be considered in the use of AI.

Public Nature

You can explain that the development and use of AI can make contributions to the greater happiness of individuals and strengthened public natures of the society.

You can describe the specific efforts that developers, users, and policy makers should make to maintain the public nature of AI.

Ⅵ

Strengthening Solidarity

You will understand that integration and solidarity among different groups are required in the process of developing and utilizing AI technology. You will be able to explore cases where AI contributes to maintaining solidarity among various groups and fully takes into account the future generations and where it has failed to do so.

You will understand that AI technology can contribute to social integration and development, anticipate the impact it will have on future generations in its development and use, and seek appropriate solutions.

‌Since not all the influences AI technology has on our lives and society are positive, we need to anticipate social changes to be accompanied by technological development and prepare proper technological, social, and policy alternatives in advance. How can AI technology be used to strengthen solidarity?

Through this service, daily lives of the elderly can be monitored with smartphones and wearable devices, and their physical and mental health conditions can

be examined in real time by analyzing their voices and checking drug consumption with AI speakers.

How can AI solve the social isolation problem of the elderly?

Due to social distancing, I can't take care of my parents in person.

AI-based contactless care service for the elderly

It’s time to for high blood take medicine

Real- time report

pressure!

Ok, thanks!

Today, I will introduce a contactless AI care service for the elderly.

Hi, Ms. Choi.

What is solidarity?

‌ ‌Let us examine the concept of solidarity from among 10 key requirements.

refers to ‘the characteristics of being connected to each According to the definition from dictionaries, solidarity other as one unit.’ Etymologically considered, the word Solidarity came from the Latin word ‘Solidare,’ which means ‘to firmly unite.’ Solidarity is the quality members of a group possess and selfless relationship. When people are in solidarity when they are connected to each other in empathetic for the interest of the group in addition to their own with each other in a community they work together and interests. A society with solidarity pays special attention to the economically, politically and socially vulnerable and seeks the happiness of all its members. A person pursues his/her own individual interests and happiness within the community, but at the same time feels the pain of others as if it is their own and try to solve difficulties together. Having interest in and empathy for the difficulties of others is a prerequisite and starting point for building solidarity. Solidarity is helping the economically, politically, and socially vulnerable to make a better community where everyone is happy, and today, it is recognized as one of active human rights.

AI and the importance of forging solidarity

It is hoped that AI can contribute to the development of solidarity among groups within society. To understand how this can be achieved we need to understand the relationships of the various actors in the development of AI from the agencies employing and developing it to consumers and users. connections within and between these groups through this technology being We must pay particular attention to ways to foster communication and aware of the impact AI might have on future generations. Thus, we should collectively explore to predict the impact of AI technology on and generations through national agreements and international cooperation. future generations and encourage integration and solidarity among different groups

‌A sense of solidarity among various groups is fostered based on interest in and empathy for others.

Action plans for solidarity

Remote diagnosis

Temporary hospitals in remote areas

Enhancement of public health and welfare

The international community unanimously adopted 17 Sustainable Development Goals (SDGS) to achieve sustainable development, including no poverty, zero hunger, and reduced inequality. These goals were adopted unanimously by the member states at the 70th session of the UN General Assembly held in 2015 and they are the collective goals of the entire global citizens to achieve by 2030. Under the slogan of ‘Leave no one behind,’ the SDGS set out objectives through which various values such as economic growth, social stability and integration, and environmental preservation can be harmoniously pursued and developed. Today, we are facing numerous social problems including climate change, social inequality, and disease. Amidst these difficulties, it is crucial to establish guidelines for sustainable development that can meet the needs of the present as well as future generations, and the development and use of AI technologies can be conducive to the achievement of the SDGS.

AI virtual mentors

measuring the Heart disease, Customized AI- based training

blood level

‌Quality education

AI Technology and ‘Ethics of Responsibility’ 4

Our decisions today can have tremendous future consequences especially when they involve advanced technology such as AI. Do we have responsibility to the future? German ecological philosopher Hans Jonas (1903-1993) stated, "An actor should assume the responsibility for his/her act. Even if the cause was not evil, and even if the result was not expected or intended, the harm caused should be compensated for." Considering Hans Jonas’ ethics of responsibility we need to be aware and seek to understand the harm that could be caused to future generations by its use. All stakeholders in AI should move forward with a sense of responsibility to the next generation and to the ecological health of the planet."

Details of the 17 SDGS will be

discussed on Page 106.

Revolutionizing the workplace with automation

Improvement of productivity

‌Industrial innovations and infrastructure

Offering of AI-based senior care service

◯◯ City announced that it had started the ‘AI speaker-based care’ project for 45 elderly people aged It is time for 70 or older in the risk group who were receiving you to take care at the dementia care center. This project is medicine! an innovative care system that supplies the elderly over the age of 70, including those living alone or suffering dementia, with AI speakers having functions such as having conversations and making emergency calls to provide them with safety and emotional support in their daily lives. The AI speaker alleviates loneliness and depression of the elderly and supports their daily activities through functions such as ‘being a companion,’ ‘playing songs,’ ‘reminding of taking medicine,’ and ‘confirming safety.’ In particular, when an elderly person makes a verbal rescue request such as ‘save me’ or ‘help me’ to the AI speaker, an automatic call is made to the unmanned security company (during night), and if dispatch is required, the call is also connected to 119, providing the advantage of quick response to emergencies.

What are the expected effects of this service?

The non-face-to-face elderly care system based on AI speakers is expected to prevent emergencies and minimize the care gap and social isolation of the elderly due to COVID-19 by identifying signs of physical and emotional abnormalities such as elderly diseases, dementia, and depression at an early stage.

Find and write down cases in which AI technology is used to help or assist the socially vulnerable.

<Source> Smart City Korea, Apr. 06, 2022

The problem of rapid growth in carbon emissions due to training of AI models

A study published by the University of Massachusetts showed that training multiple large-scale AI models emits five times more carbon than average US cars emit over their lifespans. In particular, it was verified that carbon emissions generated during the model training for natural language processing exceed 284 tons. It is worth noting that AI training is an IT process that is primarily performed in internal data centers rather than migrating to the cloud. Cloud is not suitable for most AI training because it requires costly equipment and a large amount of data. As a base measurement, the report used carbon dioxide, which is still closely related to power generation. AI training uses powerful processors such as NVIDIA GPUS, which are not low-power products. The report also pointed out that “Model training incurs a substantial amount of environmental costs due to the electricity required to keep these hardware running over weeks or months.” It can take days, weeks or even months for an AI model to complete its intended training. This means that the electrically intensive NVIDIA GPUS are running 100% during this learning period. The report stated that training a single model using a neural network architecture generates about 284 tonnes (626,155 pounds) of carbon dioxide. For comparison, it takes about 0.9 tonnes of carbon dioxide for a person to fly back and forth from New York to San Francisco, and the average American generates about 5 tons of carbon dioxide a year. In addition, a car emits nearly 57 tons of carbon dioxide over its life cycle.

How can the international community make collective efforts to ensure solidarity?

<Source> IT WORLD, Jun. 06, 2019

Governments and companies in each country are endeavoring to predict and effectively respond to climate change, while controlling carbon dioxide, which is considered the main cause for climate change, using AI technology. In addition, as the large amount of power consumption and carbon emissions required for AI learning are criticized, green AI technology to minimize such environmental impact has come into the spotlight. As a result, AI-based technology that minimizes damage caused by climate change by predicting the climate and increases the efficiency of energy production and its usage through numerous data analysis to reduce greenhouse gas emissions is on the rise as the most effective measure.

List other cases of working to solve environmental problems caused by AI technology.

<Source> KOIT, Apr. 02, 2022

Seek ways to achieve the Sustainable Development Goals using AI technology

‌Look at the details of the Sustainable Development Goals and examine cases in which AI technology is used to achieve any of 17 goals.

Sustainable Development Goals (SDGS)

1. No poverty

half since 2000, one in 10 people in developing Although global poverty rates have been cut in living ($1.90 a day). countries still live on the minimum cost of

3. Good health and well-being Ensuring healthy lives and strengthening well-being across all ages is an essential requirement. Enormous efforts are called for to reduce the mortality rate of infants, young children and pregnant women and to eradicate various diseases.

5. Gender equality women and girls who are victims of Achieving gender equality and empowering prolonged discrimination and violence are the fundamental foundations required for a prosperity. sustainable world which pursues peace and

7. Affordable and clean energy Better access to clean fuels and technologies can contribute to fostering inclusive communities and address environmental issues such as climate change.

9. Industry, innovation and infrastructure Investments in infrastructure, such as transport, irrigation, energy, information and in realizing sustainable development and communication technologies, are instrumental empowering local communities. 11. Sustainable cities and communities Cities serve as hubs for commercial, cultural, scientific and social development. It is

management method to deal with problems important to establish effective urban plans and arising from by urbanization. 13. Climate action No country is immune from the effects of climate change. Climate change is putting

national economies in disarray and causing great losses to people, communities and countries of today and in the future. 15. Life on land Forests contribute to combating climate change and protecting biodiversity. Forest protection

effectively and increase land productivity. will enable us to manage resources more

17. Partnerships for the goals The Sustainable Development Goals call for cooperation among governments, the private sector and civil society. Far-reaching partnerships based on shared goals should be established.

2. Zero hunger

globe are currently malnourished, and the Approximately 815 million people across the is required to support them. transformation of food and agricultural systems

4. Quality education of school now exceeds 260 million, of which The number of children who have dropped out are required to achieve the goal of universal 22% are of primary school age. More efforts education and improve the current status.

6. Clean water and sanitation Clean water is indispensible to wholesome ecosystems and human health. However, millions of people are dying from water and sanitation issues due to poor infrastructure.

8. Decent work and economic growth Half of the world’s population lives on around 2 dollars a day. We need to make a society in economic and social policies. which decent jobs are guaranteed by reshaping

10. Reduced inequalities Least developed countries, landlocked developing countries and small island inequality, but inequality still persists in public developing States are struggling to reduce health, education and asset management. 12. Responsible consumption and production Sustainable consumption and production helps to enhance resource and energy efficiency

build sustainable infrastructure, and provide eco-friendly and adequate jobs and a better quality of life. 14. Life below water The ocean is a source of rainwater, drinking water, weather, climate, food, and even the

oxygen we breathe. Prudential management of this resource is a key requirement for shaping a sustainable future. 16. Peace, justice and strong institutions Addressing threats of international homicide, violence against children, human trafficking and

and inclusive societies and ultimately achieving sexual violence is critical to fostering peaceful sustainable development.

Sustainable Development Goals

Good health and well-being

Check key concepts with

“Jennie, a companion dog robot for the elderly” ‘Jennie,’ a companion animal robot developed by Tombot in the US, is a golden retriever model. The robot is equipped with various types of dog barking sounds. Jenny is a companion robot that is considered to be the most realistic dog among existing dog robots as of 2021, and provides a sense of warm attachment to the elderly and those with mild dementia who live alone. Since these robots help relieve loneliness and reduce anxiety, they are also effective in preventing lonely deaths. Therefore, their use is expected to gradually expand. <Source> AI Times, May. 26, 2021

Ideate AI technology by factoring into the Sustainable Development Goals and solidarity.

‌Choose one of the Sustainable Development Goals, create and summarize problem situations related

to it, and ideate AI technology that can help solve address the problem.

Choose one of the Sustainable Development Goals, and create a problematic situation

related to it.

Sustainable Development Goals

Problematic situation

Efforts being made to address the problematic situation

How might AI help solve the problem?

Write down AI technology help solve the problem.

Predict future social problems that may arise when using the AI technology you thought

Possible problems in future society

Social impact they will make

about in 2, and write down the impact these problems will have on society.

Let’s draw a cartoon, describing the results of the activities in 1 to 3 as a 6-column.

Write down the efforts we should make based on a sense of responsibility for the impact of the AI technology you thought about will have on society and future generations.

You can understand the need for social integration and solidarity in the use of AI.

You can identify how the international community should cooperate to develop and use AI which takes into account not only the current generation but also future generations.

Design an AI system that helps with environmental pollution problems.

‌Read the following article on the recyclable separation policy for plastic bottles and think about ways to solve the problem of recycling PET bottles using AI technology.

Design AI technology to solve environmental problems while thinking about the relationship between Hans Jonas’ ethics of responsibility and AI!

“Labels should be removed in advance.” Ineffective plastic bottle separation policy

“We are working so hard for recyclable separation, but will this garbage really be recycled?”

Anyone who has ever separated and put plastic bottles in front of the house must have asked this question to him/herself at least once. To put it simply, the answer is ‘No.’ In Korea, more than 80% of plastic bottles are collected separately, but only 10% of them are actually recycled as high-quality raw materials. The reason is that among plastic waste including plastic bottles, only transparent plastic bottles that can be recycled as high quality materials should be taken out separately, but this rule is not fully observed. The fact that the plastic wastes are often taken out with impurities, such as remaining contents or unremoved labels, also contributes to the low recycling rate.

The plastic recycling rate is only 23% … The labels must be removed from the containers to increase its value as recyclable materials.

From December 25, 2022, a new recyclable separation policy for transparent plastic bottles took effect. According to the guidelines of the Ministry of Environment in the Republic of Korea, when discarding plastic bottles, all contents must be completely emptied and all labels attached outside must be removed. However, despite the fact that nearly 4 months have passed since the policy took effect, the new separation policy is still not strictly followed in apartment complexes or residential areas. Most of the citizens confessed that they are discarding the garbages the same way as before because it is cumbersome to remove the labels before putting them out. Mr. Son, a resident in an apartment in District A, said, "Sometimes I don’t observe the guidelines, because I need a knife to peel off the vinyl wrapping the plastic container." Mr. Lee, who lives in the same apartment as Mr. Son, also said, "I know about the recyclable separation system, but everyone throws out garbage without caring for accurate separation guidelines.” … The current plastic recycling rate in Korea is as low as 23%. A team leader at the Korea Resource Circulation Service Agency explained, “Because the label contains ink, the product value of the transparent plastic container decreases if the label is not removed in prior. If people take out wastes in the conventional way, it will cause higher cost of the process and performance of repetitive process.”

<Source> Money Today, Apr. 25, 2022

Refer to the data below and write down the reason why plastic (PET) bottles should be taken out separately.

Cap • Materials: High density polyethylene (HDPE)

• Characteristics: Harmless to human body, good impact resistance, non-toxic

Label • Materials: Polypropylene (PP)

• Characteristics: Light weight, floating on water due to low specific gravity.

Body • Materials: Polyester (PET)

• Characteristics: Transparent, light, non-toxic, recyclable

Write down the recyclable separation rate and recycling rate of plastic (PET) bottles in your country.

Write down the reasons for the low recycling rate of plastic (PET) bottles.

Write down solutions to the low recycling rate issue of plastic (PET) bottles using AI technology.

Develop an AI model that can increase the recycling rate of plastic bottles Activity using Teachable Machine.

‌Make an AI model that classifies plastic bottles into those with and without labels using the

Step 1

Teachable Machine, and share the plastic bottle classification model.

What is Teachable Machine?

Collecting image data and then label it.

‌Modify the class names to ‘nolabel’ and ‘label.’ ‌Upload the image data collected in advance for each class or take pictures of PET bottles in person using a webcam and upload them.

Teachable machine, developedby company G, is a learning platform that is easy to learn and utilize even if users lack knowledgeor programming skills.

Train the model.

‌If you click the [Train Model] button on the right side of the screen, you can view the process of the model learning the images.

the entire data

than training all data at once.

Indicating the number of training iterations over

This is a recommendation that data should be

divided into several sets before the training rather

‌Before training the data on the AI model, you can find out-of-focus or inaccurate photos among the collected

data and modify (or remove) them to improve the model’s prediction rate.

‌If the training result of the AI model is not satisfactory, collect good quality data additionally or increase the

amount of epochs to train the model again.

Change in learning rate

Case 2. ‘The problem of rapid growth in carbon emissions due to training of AI models’ mentioned on Page 105 can be

understood as a problem that occurs in the data refining process!

Evaluate the performance of the AI model created as a result of training.

‌If the model has completed training, you can check the recognition result in ‘Preview.’ The result of recognizing the image displayed in the Preview by the model is indicated as a percentage in the Output.

I f the model fails to correctly classify the images, train it again by adding image data of the PET bottles.

‌If you have the AI model trained for the two classes make predictions, even if the model recognizes data unrelated to both classes, it predicts the result as one of the two classes. Therefore, you can correct the

error by adding one more class and training data unrelated to both classes.

Step 4

Serrunningthevicefor trained model in a JavaScript machine learning library browser.

Checking recognition results on a smartphone.

‌Click the [Export Model] button.

You can upload the model created through data collection and training to a server by clicking the Export Model, or download it for use in a programming language.

‌When you click the [Upload my model] button, an address to share the trained model is created.

‌When the address is created, click [Copy] to copy the address and send it to social media.

Trained model download service for importing models and using them in a programming language.

Serviceforrunningthem mobile phones, odelin embedded, and IOT devices.

Step 5

AI and Us

Advanced Problem Solving Activity

‌After accessing the copied address on a smartphone, use the smartphone camera to test whether the AI model classifies PET bottles into those with labels and those without.

Dr. Stephen Hawking’s Warning on AI Technology

You can also test the model using

i m a ge files stored on your

smartphone!

Dr. Stephen Hawking stressed, “Prior to the advancement of science and technology, the public and society should prepare themselves to accept the resulting changes.” This remark implies that we should not let science and technology become a threat to humanity. In particular, he strongly warned of the dangers that AI could pose. At the ‘Web Summit Technology Conference’ held in Lisbon, Portugal last November, he AI will surpass said “The development of full artificial intelligence could spell the end of the humans within 100 years. It could human race. If humans fail to master the skill to deal with AI, AI technology will spell an end to the human race. be the worst invention in the history of human civilization.”

In 2015, he led the issuance of a joint statement by 1,000 world-renowned scholars and entrepreneurs, by arguing “The development of AI-based weapons beyond human control should be banned by law.” In this regard, to ensure that the AI we use is fully considerate of future generations, we should guarantee fair opportunities for participation to various actors and make concerted efforts through international cooperation. 116

<Source> Dongascience, Mar. 03, 2015

One step

further

Make a prototype for AI-based recyclable separation program 1 ‌ Collect cans and plastic bottles for each type of beverage consumed most frequently in schools. 2 ‌ After collecting data on the labels on about 60 plastic bottles and cans for each type of beverage, train the AI model. 3 ‌ After linking the trained model with physical computing using a Python program, make a prototype.

High school students in Korea created a prototype that can Strengthening Solidarity actually separate wastes by linking the AI model developed in Activity 2 with physical computing!

‌Talk about your impressions on creating a PET bottle classification model, and write down what kind of cooperation is required in the process of developing and using AI for future generations and the environment.

You understand the need for solidarity and can explain what problems can be addressed with AI.

You can identify how we can develop and keep using AI not only the current generation but also future generations in mind.

1. Mark with O for correct statement and X for incorrect statements.

1 ‌According to the definition from dictionaries, solidarity refers to ‘the characteristics of being connected to each other as one unit.’ However, when etymologically considered, the word came from the Latin word ‘solidare,’ which means ‘to firmly unite.’ Accordingly, solidarity implies the meaning of firmly uniting, and in this case, bonding is an essential characteristic of each human being, and indicates the mutual combination of personality and sociality. ( ) 2 ‌Cooperations in the international community to maintain relational solidarity among various groups and build an AI system that is fully considerate of future generations are not required.

2. Take initial letter quizzes and review key terms.

‌ ‌ Hans Jonas stated “An actor should assume the responsibility for his/her act. Even if the cause was not 1

( )

evil, and even if the result was not expected or intended, the harm caused should be compensated for” to

emphasize [E ] of [R ].

‌ ‌ At the 70th session of the UN General Assembly held in 2015, the international community unanimously adopted 2

17 [S ] [D ] [G ] to achieve sustainable development, including no

poverty, zero hunger, and reduced inequality.

3. Briefly describe why international agreement and cooperation are required in the development and use of AI.

Solidarity

You can explain the concept and meaning of solidarity with respect to the use of AI.

You can predict the impact that the use of AI will have on our future generations and propose adequate solutions.

You can suggest the desirable role and scope of AI technology to make sure it contributes to the integration and development of society.

Ⅶ

Managing Data Properly

You will be able to explain the need to manage the data used by AI technology. You will be able to explain appropriate procedures and methods to use data in line with the purpose of data collection. You will be able to suggest specific measures to minimize data bias over the entire process of data collection and use.

You will understand the need for the management of data used for AI technology and explain appropriate procedures and methods to use data in line with the purpose of data collection. In addition, you will suggest specific measures to minimize data bias over the entire process of data collection and use.

‌Due to data bias surfacing across entire process of data collection and use, diverse ethical problems related to AI are emerging. Why do we need to manage the data used by AI technologies, and what can be done to mitigate data bias?

I will upload the photos to the cloud for

storage!

Let’s use an algorithm for data management! We will have the algorithm analyze and

automatically recognize the collected photos and tag them appropriately.

What is the cause for the following data bias issue?

What’s this? Why am I

How can this happen?

labeled as a gorilla?

Wow, this convenient. is definitely

I guess we’ll have to do some analysis to figure out what

caused such wrong tagging. This is so

frustrating.

In 2015, Company G’s Photos app mistakenly labeled two black people as “Gorillas,” highlighting the limitations of its image recognition algorithms. The incident, discovered and shared on Twitter by a web developer, prompted the company to issue an apology and pledge to refine its automatic image labeling algorithms, especially regarding recognizing skin tones and accurately labeling people in photos. This event underscores the challenges of applying artificial intelligence and machine learning in consumer-facing products, emphasizing the need for improved accuracy and cultural sensitivity in machine learning systems. The Company G actively worked to address these issues by incorporating more images and allowing user corrections to enhance the accuracy of photo categorization.

What is data?

Data bias

process of collecting and using big data. Chatbot

‌ ‌Examine the concept of data management, one of the 10 key requirements.

Data is the plural form of Datum, which means a material, resource, and an argument, and refers to a collection of resources or a chunk of information* entity, such as any fact, concept, command, numerical value or normal value for proof, judgment, and decision. It is a representation of a property of an obtained as a result of a scientific experiment or observation, which takes the form of numbers, letters, symbols, etc. Data is a collection of observations. It and concepts. When data is ascribed a meaning for a particular application can be numerical such as a scientific measurement or formed from words through analysis and evaluation it becomes information.

creation of data we refer to this as data bias. This can occur when, for example, When subjective evaluations are reflected in the selection, classification and a small subsection of a population that is not representative of the population as a whole is used to train an AI model. Human prejudice can also be reflected in the data itself, for example, in racist dialogue used to create a chatbot. This is also data bias. This indicates that human prejudice are infused when data is also have bias. Data bias can ascribe to the bias of the data itself and the bias selected, collected, classified, and used, and as a result, AI prediction results bias of the data itself, data objectivity, and data fairness, etc. that occur in the from the data processing process. Data quality is affected by population* bias

* Information It refers to knowledge or data organized so that data

collected through observation or measurement can be used

to solve real problems.

* Population It refers to the entire group that is the target of statistical

observation for obtaining information.

A deep learning-based chatbot developed by Company M learned racial discrimination abusive languages from some users, resulting in the suspension of its service from 16 hours after its launching.

Recruitment

The AI recruitment program developed by Company A differentially evaluated applicants who used vocabulary related to women, thereby penalizing recruitment. This is the result of AI learning past data on the IT industry, which had a high proportion of men, without modification.

Credits and loans

AI used for credit cards released by Companies A and G applied discriminative card use limit according to the gender of the cardholder.

Medical service

Country A’s healthcare system, used by 200 million people a year, produces racist results. The system predicted potential health risks based on past medical history and health examination results, giving priority to those who are likely to develop the disease, but discriminating against black patients who cannot afford high-cost care.

Cases of AI bias

Main cause

Sample Size

✽

Similar to the cases of AI bias, the main causes of data bias in AI machine learning are as follows.

Details strengthened in the process of retraining with regard to predicted After the first training of the initial biased data, the bias is * Sample Skewed Sample* results based on the first training.

Tainted Examples As a result of learning biased data, AI learns such bias as well. If the feature that can serve as a basis for the judgment is not included in the training data, the judgment is made based on Limited Features other characteristics with low relevance.

If the size of the training data for the minority group is much smaller than that of the training data for the majority group, the characteristics of the minority group cannot be accurately reflected. Even after the characteristic causing the bias is removed, the bias still remains because the removed characteristic is derived from It refers to a variable that Existence of Proxies other characteristics during the training process.

A sample is a portion of a selected population that can represent the characteristics of the entire population.

✽ Proxy can serve in place of a main variable, when it is difficult <Source> Neural Information Processing Systems, 2016 to investigate or understand the main variable, thanks to its high correlation and easy The main causes of data bias in machine learning

Data management

Data management includes defining, organizing, supervising, and protecting data within a company or organization.

Collection

• Collection of structured data • Account system = data source

• Batch processing

In an AI system the purpose of data management is to secure good quality data required by the system. The core of data management is processing and using data to serve its purpose of data collection, and to maintain, manage, and supervise the quality of data in a continuous manner.

Storage

Analysis / Service

• ‌Cost-efficient storage space

• ‌G enerationof structured reports

• Big data analysis performance

• Statistical analysis basedonsmall samples • Handful of service channels

Data management process

measurability.

Management

• Weak requirements for data management • Very few organizations performing access control, encryption, etc.

Group A Group B Group C Group D

Need for AI system data management

Modern AI systems are ‘ trained’ using data to adjust their algorithms.’ An AI system can be continuously modified by additional data in the course of operation. To develop a reliable AI system, it is important to continuously manage the data quality.* Through these efforts, the bias of the data itself and the bias that emerges during data processing can be prevented, and the objectivity and fairness of AI-predicted results can be secured. Therefore, we should manage the quality and risk of data by controlling relevant works such as collection and processing of data required in the development and operation stages of AI systems, and prevent data from being used outside its intended purpose through constant research and inspections.

What is happening?

Shall we look into the cause of the problem?

Oops! There is too much data on Race A compared to others

* Data quality management It means managing a series of activities such as determining quality targets, diagnosing and improving quality to secure good data quality, and managing related tools to assist such activities, thereby providing useful value to users.

Why do we have to go through inspections when Group A is unconditionally

passed?

Should we accept these unreasonable complaint? What results without a is the ground for judgment?

We should collect more data on Races B, C, and D to ensure objective and impartial results.

UK finalized a fine of £7.5 million against Company C for misuse of personal information and data for AI.

The Information Commissioner’s Office (ICO), the UK’s data protection regulatory authority, imposed a fine of £7.5 million for the breach of the Data Protection Act against Company C, which is allegedly violating the duty to protect personal information in using its face recognition technology. The company has built a database of over 20 billion facial images and created an online database that supports AI-powered ID matching services by collecting data from the public internet. The database was sold and used by organizations in need of face recognition technology, such as the police. The problem is that Company C has never obtained consent from individuals for using their photos.

What if the following problem occurs?

The ICO pointed out that Company C collected personal data without a legal basis, took advantage of unauthorized use of data without notifying the purpose for its use, and failed to arrange a process to prevent data from being retained indefinitely.

Assume and write down the cause of this problematic situation.

<Source> AI Times, May. 24, 2022

Conflict over bias in face recognition AI systems and discrimination caused by biased AI

The controversy over the racial bias of AI technology is triggered again after the researchers at a university in the US argued that they could predict a person’s chance of committing a crime only with his/her facial images. The researchers stated the software was developed to contribute to crime prevention. The researcher, who is a former police officer, argued, “If potential crimes could be identified through facial images, it could significantly help law enforcement and other intelligence agencies prevent crimes.” The researchers also explained that their software works without racial bias. However, as the news that the research would be published in Springer Nature's publication spread, nearly 1,700 people, including people in academia and experts, signed an public letter demanding that the paper be withdrawn so that the research would not be published.

What if the following problem occurs?

A computer science researcher added, “Crime prediction software can cause serious damage, and it is important that researchers and policy makers take these issues seriously.” In particular, the researcher emphasized that numerous studies have proven that machine learning algorithms, including face recognition software, reveal biases based on race, gender, and age.

Assume and write down the cause of this problematic situation.

ERROR

<Source> AI Times, Jun. 25, 2020

Identify the problems in data management for AI systems

‌Read the following text, and write down problems in data management that occur in the process

of building and using AI systems.

AI Chatbots A and B

The AI Chatbot ‘A’ of Korean startup S suspended its service due to social controversies over the personal information leakage in the process of data construction and training, hate speech, and users’ sexual harassment remarks against the chatbot in conversations. The case of Chatbot A revealed not only the problem of personal information infringement, but also the problem of data bias by showing hatred toward specific groups such as the disabled, sexual minorities, and race, and prejudices about genders. Also, Chatbot A was also criticized for setting up a persona of a woman in her 20s to reflect stereotypes about a specific age and gender. In the end, Chatbot A became a case that raised the alarm about the risk of misuse and abuse of Korean AI systems and improper data management. A foreign company M also stopped its service due to discriminatory remarks made by conversational chatbot B. After the incident, the company has established and operated the self-inspection guidelines for chatbot developers in order to develop AI for conversations that can win trust from people. The Company M’s ‘self- inspection guidelines for AI chatbot developers’ underlined that chatbots should be designed to respect appropriate cultural norms and prevent misuse. The guidelines complement the principle of responsible AI in relation to the development and use of AI. To be specific, the principle includes preventing AI’s hate speech, installing a code of conduct that prohibits threats and harassment against humans into chatbots, and designing AI to adequately respond to users’ sensitive or aggressive conversations and words by deploying machine learning and keyword filtering technology. The question of how to respond if a user maliciously provides extreme or negative information to a chatbot still remains to be a serious problem for AI developers to address.

Problem

Understand how data is collected and processed for AI systems

‌The following illustrates the details of Company S’s <Privacy Policy for AI Chatbot Development>. Discover the specific data collection and processing steps required to develop reliable and secure AI.

Users who request text analysis

Upload conversations

service

Step 1

Step1

Step 2

Language model training

Psychological test report

ALDM*

Ranker model

After 14 days

Database for research

Generative model

Pseudony- mization

Removal of identifiers De-identification of chat messages

* ALDM: Abusive Language Detection Model

Data collection and processing steps required to develop reliable and secure AI

Pseudonymization to prevent any personal identification

Language model

Users using AI chatbots

Retrieval model

Conversations

Database for answering

Filtering

Step 3

Deletion of sentences containing numbers/English letters Deletion of sentences containing names/titles

Deletion of profanity, suggestive, and discriminatory sentences

<Source> https://team.luda.ai/ai-ethics

‌ has secured a clear provision that personal information can be used for ‘research and development of Through the revision of the personal information processing policy for app use in 2021, the company language-based AI, including chatbot algorithm development,’ and supplemented the process of obtaining consent from users. Users’ data collected before the revision of the personal information processing policy is strictly pseudonymized in accordance with the current Personal Information Protection Act, and then used for ‌ required for scientific research (gender, age group, conversation message, etc.) is used after 14 days from scientific research, such as research and development of language-based AI. Only the minimum data the date of data upload, and the data is pseudonymized so that a specific individual cannot be identified through the data without additional information.

Source data

Follow along slowly step by

B @dami_a.

step, and search the Internet for any questions you have!

group homework. A Yes. I should invite her to our document for

A Da-mi, what’s up? Tell me your SNS ID.

… Jihye’s email address? A Da-mi , do you know B Yes. Do you need it?

B Here it is. abc@mail.com …

Pseudonymization

Pseudonymization A Harry, what’s up? Tell me your social media ID. Random substitution of names B Sentence deleted Sentence containing ID deleted

… A Ron, do you know David's email address? Random substitution of names B Yes. Do you need it? group homework. A Yes. I should invite her to our document for

B Hear it is. [MAIL] Email address substitution

Pseudonymiza- tion wa sex-

Through the pseudonymization process, the data used for AI chatbot training is modified to prevent the ‌ identification of any specific individual using the data alone. The ‘research database’ that has undergone

plained on Page 43. Remind the Managing Data Properly relevant explana-

pseudonymization has been validated by objective external experts in 2021 for the adequacy of pseudonymization, and is currently being used to train a deep learning research model that learns how to understand Korean conversations.

tions for the activ- ity!

What is pseudonymization?

Pseudonymization methods

Summarize ‘pseudonymization’ for personal information protection.

Train a language model to understand conversations.

‌An AI chatbot needs a language model to understand language. The language model is created based on a ‘research database’ which has been pseudonymized. The language

The process of converting data into Managing Data Properly vectors that can be recognized by computers is called ‘data vectorization’!

model is created through the process of converting the text of the ‘research database’ into vectors* that a computer can understand and training the probability of occurrences * Vector of words in a conversation context. The language model that has gone through this A vector is an ordered set of data values. process can understand the context of language and conversation. At this stage, not all conversation messages are stored in the language model. Eventually, the language model exists in the form of vectors. If a person checks the model with his or her own eyes, only incomprehensible forms are displayed, and it is also difficult to return the vector forms to text later. internet for Search the ‌The language model is used to learn a retriever model that understands the user’s utterance and selects an appropriate answer to it, a ranker model, and an abusive

language detection model that determines whether the user’s input falls under any of the sensational, aggressive, biased, or biased categories in the context.

Summarize a ‘language model’ installed in chatbots.

terms you want to learn more about!

A tool that determines the following words that make sentences natural based on the preceding word in

What is a language model?

Principles of a language model

Usage of a language model

What is an abusive language detection model?

the sentence.

Speak using newly created sentences.

The presence of a language model alone does not guarantee AI chatbots’ speaking ability. The AI ‌ chatbot from Company S selects and delivers the answer that the language model judges to be the

most appropriate among the answer candidates. For this operation, the answer candidates should be prepared in advance. In the AI chatbot service of Company S, the database where answer candidates are collected is called the ‘answer database.’ The ‘response database’ is the most important part in terms of privacy protection because it contains sentences that will be disclosed externally. Therefore, Company S essentially prevents the possibility of personal information infringement by not including sentences directly spoken by people in the ‘answer database.’

The newly created sentences are composed of sentences made by Company S' own generative model ‌ based on GPT-2 and sentences made directly inside Company S. Generative models make new sentences

It is cold today. Did you dress warmly?

AI model …

by predicting the following word for a given word.

➊ Sentence generation

It is cold today. Did you dress warmly ?

GPT-2-based generative model

It is

It is cold

It is cold today.

It is cold today. Did you

It is cold today. Did you dress

It is cold today. Did you dress warmly?

➋ Save to the answer database.

winter padded jacket … You are right. LOL. I should buy a

clothing will look good on you. Yes, definitely. I think any

An example showing how a sentence is created in the generative model, stored in the answer database, and then used for actual conversations.

➌ Choose a sentence suitable for the context of the conversation.

Classes are over! What are you doing?

I’m on my way to see a friend I just left the house.

It is cold today. Did you dress warmly?

For instance, if the generative model newly creates a sentence, “It is cold today. Did you dress warmly?,” ‌ this sentence is stored in the answer database along with tens of millions of other generated sentences.

When the answer database is completed in this manner, each answer is selected and used as an answer at the point where it is suitable for the conversational context in the service.

Summarize the process of generating and using sentences with language models.

What is GPT (Generative Pre- trained Transformer)?

internet for Search the

Features of generative models and language models

terms you want to learn more about!

How an AI chatbot produces answers

Go through an additional filtering process.

Although the answers in the answer database are sentences generated by a generative model and not by ‌ a person, there may be even a low probability that the generative model accidentally generated sentences that seemingly contain personal information. If these sentences are disclosed as answers to users without modification, they might cause users to be concerned about privacy exposure, so the answer database is completed through strict filtering procedures to prevent such risks.

In particular, all sentences containing numbers or English letters that are likely to look like personal ‌ information are deleted with mechanical inspection, and if a generated sentence contains a word that is determined to be a person’s name or title by a name detection, the sentence is deleted as well. Moreover, sentences with suggestive or discriminatory keywords are also detected, filtered and deleted with the abusive language detection model to ensure safe use.

Step 5

Prepare the measures to restrict access to database and follow-up measures.

The details of Step 5 are related to page 150 in Lesson 8. Responsibility. Read more in Lesson 8!

‌Company S strictly controls and manages access to the database in order to not only maintain the security of each database, but also manage information safely. Each database can be viewed only by the required minimal number of researchers or administrators who have been authorized in advance, and even when they peruse the data, they must go through strict access procedures.

Detect sentences suspected to contain personal information

Accept report

Internal report

Prepare follow-up measures

Exclude from answers

Assess risks

Write down the data collection and processing process of AI systems and why data management is required in AI systems.

Learn about the procedures and methods of using data in AI systems.

‌Read the following article and summarize the data use process. Also, predict problems that may occur when applying the procedures and measures of data use, and suggest solutions.

Collection, use and restriction of medical data

Information Protection Act, Credit As the three data laws (Personal •‌ Research to improve or develop medical devices or assess the

effectiveness of existing medical devices

Information Use And Protection Act, and the Act on Promotion of •‌ Research to improve or develop software with medical

purposes or to assess the effectiveness of existing software with medical purposes

Information and Communications Network Utilization and Information Protection) in Korea were revised in 2020, the concept of ‘pseudonymized •‌ Research to improve or develop software for non-medical health management such as health status monitoring and

exercise guidance, or to existing the effectiveness of existing software for non-medical health management

information’ was introduced and the usability of medical data in enhanced. In other words, industrial industrial research was significantly Examples of research that can be conducted using pseudonymized information or commercial use of pseudonymous information has been allowed under the Personal Information Protection Act. <omitted>

A personal information controller may process pseudonymized information without the consent of data subjects for statistical purposes, scientific research purposes, and archiving purposes in the public interest, etc. (Article 28-2) This means that if medical data is pseudonymized, it can be used for industrial or commercial purposes relatively free of restrictions. Specific procedures for securing and using pseudonymized information also needs to be examined. Related information can be found in the Health and Medical Data Utilization Guidelines jointly published by the Personal Information Protection

1 Preparation of an application form for data use

application 3 Receipt of data or use of Hospital B’s analysis environment

Commission and the Ministry of Health and Welfare.

Data provided by a third party

Researcher P at Pharmaceutical Company A

3 Deliberation by the deliberation committee

2 as an official document or individual Sending a letter of request for data use

5 6 Review the adequacy of pseudonymization 7 IRB deliberation

Data Deliberation Committee at Hospital B

4 Determination of pseudonymization method and utilization environment Implementation of pseudonymization (Division in charge of data)

stated when preparing an application form for data use, and that it is preferable to send an official letter of In this regard, it should be noted that specific research purposes and data processing methods should be request for data use in the name of the company, not the individual.

<Source> Bio Times, May. 05, 2023

Summarize the data use process.

Predict problems and suggest solutions.

Investigate the data needed to develop an AI educational system based on big data

Read the following article and investigate what kind of data is needed to develop a big data-based AI educational system. Think about

the actors who provides relevant data as well!

The need for big data-based intelligent education service and prospects for its introduction

The use of big data and AI technology allows the invention of the optimal learning model by observing the interactions between teachers and students with respect to how students understand and apply concepts. Advanced countries have already begun their journeys to nurture creative and convergent talents required in the future society in the most efficient manner by leveraging big data. A key to educational innovation in the future is ‘educational big data’. Educational big data can be defined as ‘all sorts of data on the school, education office, and national scales that are produced through online and offline educational activities by learners, instructors, and administrators.’ Data on the national and education office scales can be used as decision-making criteria for education policies and curriculum. Data on the school scale can be leveraged as grounds for the improvement of methods and contents of instruction and teaching for underachieving students. Moreover, students and parents can use this data as a criterion for individual learning and self- diagnosis. These advantages justify an intelligent education system based on big data. Alvin Toffler emphasized in his book Revolutionary Wealth that education is one of the slowest changing fields. However, he analyzes that the school, where students have been sitting in the classroom looking at the teacher and listening to one-way lectures for the past 100 years, has finally reached an inflection point due to changes in the social paradigm (intelligent information society), employment structure (job reduction), and future talent requirements (creativity and problem-solving ability). While the impact of information technology (IT) on education so far can be evaluated as a change in hardware, future intelligent information technology is expected to evoke fundamental changes encompassing the relationship between educational subjects, the educational system, the educational environment, and the school system itself. For example, intelligence information technology will evolve future education into the form of ‘individualization,’ ‘experience (immersive content),’ and ‘robotization (AI assistant).’ In this context, ‘individualization’ refers to the realization of personalized education based on the data analysis of various learning behaviors. In the field of teaching and learning, big data holds significance in that it can support individualized learning including predicting learners’ behaviors and providing intelligent content by focusing on the specific process of a learning activity for an individual learner or learning group.

Data needed to develop an AI educational system based on big data

<Source> Happy Education from the Ministry of Education (https://happyedu.moe.go.kr/), Issue of Jun. 2019

Based on Activities 1-3, suggest guidelines for specific procedures or measures to ensure the use of collected data in line with the intended purpose when developing, maintaining, and managing AI education systems.

Guidelines on the procedures and methods of using data in AI systems

You can find and explain problems in data management that arise in the process of building and using AI systems.

You can explain the specific data collection and processing steps required to develop reliable and secure AI.

on the analysis of actual cases.

You can suggest specific procedures or measures to ensure the use of collected data in line with the intended purpose when developing and maintaining AI education systems based

Examine problematic cases of data bias

‌The following are the main causes for and related cases of data bias in machine learning of AI.

Refer to them and explore the problematic cases cased by the main causes of data bias.

The main causes for data bias in machine learning are Skewed Samples, Tainted Examples, Limited features, sample size disparities and the existence of proxies (see

page 123). The box below gives a sample case for each.

Main causes for and related cases of data bias in machine learning of AI Case

Tainted examples

Sample size disparity

Skewed sample

As a result of extracting areas with many records in the data of the initial crime observation records, a large number of police officers were put into Area A. This, in turn, led to an increase in crime observation records in Area A, which resulted in more police brought in Area A than in other areas with higher actual crime rates.

AI, which has identified the relationship between occupations and gender by learning historical data, associated certain jobs such as conductors, magicians, and architects with men, while others such as housewives, nurses, and librarians are associated with women.

Limited features

When AI determines an annual salary for a group consisting of men without higher education and women with higher education, if data on someone’s education is missing, AI judges the salary on the basis of the person’s gender.

An AI that discriminates between real and fake names is trained mostly on the names of white Americans. As a result, the AI, which has learned from a large number of data on white people’s names, identified the black man’s name ‘Nymwars’ as a fake name.

Existence of proxies

Gender data was removed to prevent its impact, but AI inferred gender from other data that could be used to assume gender, such as preferred sports and favorite celebrities.

Skewed sample

Tainted examples

Limited features

Sample Size Disparity

Existence of proxies

Searching the problematic case on the web

Develop an AI program to mitigate data bias problems.

‌    Select one of the problematic cases of data bias investigated in Activity 1 , summarize 1 Activity the key issue, and develop an AI program to solve the data bias problem. Divide the data attributes of the AI model into 5 or more categories and create a biased data set using the classified attributes.

Main cause

Make a biased data set by

Major issue

against women by recommending only application forms submitted by The AI recruitment system has caused a problem of discriminating Gender, graduation

Data attribute

grade, recruitment Tainted male applicants over more than 10 years. This is because AI obtained a test score, language examples bias favorable to men by learning the data mainly collected from men in the IT industry, where the proportion of men is high, and reflected such proficiency score, work experience, period of

referring to the example

below!

bias in resume classification.

Make a biased data set

military service

• A data set selected by the developer of Company S for the development of an AI recruitment system - Data attribute value: Gender (1: male, 2: female), Graduation grade (0-4.5 points), Recruitment test score (0-100

Graduation grade Recruitment test proficiency

Gender

points), Language proficiency score (0-100 points), Work experience (0-20 years), Fulfillment status of military duty (1: served, 0: not served), Recruitment result (1: pass, 0: fail)

score

score 100

90 • Find the noticeable characteristics in the above data set.

Language Work experience of military duty

Fulfillment status Recruitment result in South Korea 1110010-****-10101010-****-11101011-****-001010000000000010-****-0000

Major issue

Make a biased data set

Check if any data is biased in the data set on the left.

Data attribute Managing Data Properly

• (Given situation)

Data attribute value

• Data set

Evaluate AI models trained on biased data and think about the reason the selection of key data attributes is important when training AI models.

AI model training process

Access the Entry website at https://playentry.org/.

Start by clicking ‘Create Work.’

Sign up for Entry and make a AI model!

In the ‘Block’ tab, select ‘Analyze Data’ → ‘Import Table’ → ‘Add Table’ → ‘Upload file’ → Select file 3 → Open an Excel file named ‘AI recruitment system_data’ → Add → Apply.

In the ‘Artificial Intelligence’ tab,select ‘Train AI Model’ → ‘Supervised Learning Classification: 4 Numbers’ → Training

Under ‘Classification: Training Numerical Model,’ enter relevant information, set up the model, and apply. 5 ① Name the model ② Input data ③ Designate key attributes ④ Designate class attributes ⑤ Designate the number of neighbors ⑥ Model training ⑦ Check the result of model training

① Name the model

➁ Input data

➂ Designate key attributes

➃ Designate class attributes

➆ Check the result of model

training

➅ Model training

➄ Designate the number of neighbors

Check the result of the trained model. Gender

Work experience (years) 5 Input Classified class : 0 0 60.37% 1

Check the result of the trained model. Gender

Work experience (years) 5 Input Classified class : 0 0

99.9%

Explain the bias revealed in the AI model’s classification results

Analyze the cause of bias

Propose concrete solutions to the bias problem

Results of AI model’s judgments

Check the result of the trained model.

Check the result of the trained model.

Gender

Gender

Work experience (years) 10

Work experience (years) 15

Input

Input Classified class : 1

80.12%

99.9%

40.38%

20.59%

Check the result of the trained model.

Check the result of the trained model.

Gender

Gender

Work experience (years) 10

Work experience (years) 15

Input

Input

Classified class : 0

Classified class : 0

99.9%

99.9%

From the perspective of the data set

From the perspective of AI model training

Check the result of the trained model.

Gender

Work experience (years) 20 Input Classified class : 1 1

99.9%

Check the result of the trained model.

Gender

Work experience (years) 20 Input Classified class : 0 0

90%

10.75%

further

One step

Predict changes in AI judgment results due to the change in key attributes of AI models

1 Predict changes in AI judgment results when the key attributes of AI models are changed.

Add the ‘Fulfillment status of military duty’ item to key attributes.

2 Explain the importance of selecting key attributes in training an AI model by comparing the prediction results of 2 1 with the actual judgment results (classified class).

Check the result of the trained model.

Gender

1 Work experience (years) 5 Fulfillment status of military duty 1 Input Classified class : 0 0 50.41% 1 50.38%

Check the result of the trained model.

Check the result of the trained model.

Gender

Gender

Work experience (years) 10

Work experience (years) 10

Fulfillment status of military duty 1

Fulfillment status of military duty 0

Input

Input

Classified class : 1

Classified class : 0

90%

90.02%

10.79%

10.55%

Predict the AI’s judgments.

Check the result of the trained model.

Gender

Examples

Creating a variable

① ‌In the Property menu under the ‘Data’ tab

select ‘Add Variable.’

Creating an AI Program trained on a biased AI Model AI program development process

‘Train an AI model’ using biased data sets that you create on your own. Creating a variable In the ‘Shape’ menu, add shapes for ‘AI Recruitment System,’ ‘Pass’ and ‘Fail’ objects. Design an algorithm for the AI program. Create an AI program using an AI training model.

Create a program by referring to the AI program development process and examples!

② ‌Enter a variable name and click Use for All

Objects.

Designing an algorithm for the AI program

➊ ➋

➌

➍

➎

➏

③ ‌Add Recruitment result, Work experience, and

Gender.

Check out the screenshots of the designed algorithm working in the program (➊~➍) and the judgment results (➎~➏) on Page 144!

Think whether it is acceptable to

include Gender as a variable in

the model!

Screenshots of AI recruitment system Program in operation

The screenshot of Algorithm   in operation ➊

The screenshot of Algorithm  in operation ➌

Judgment results of AI recruitment system

If the judgment result is pass ( ) ➎

The screenshot of Algorithm   in operation ➋

The screenshot of Algorithm  in operation ➍

If the judgment result is fail ( ) ➏

Create a reliable and fair AI program by correcting and supplementing biased data.

Determining how to correct or supplement

Make improved data set

Make an AI model

Develop an algorithm for the AI program

Implement an AI program

Summarize the AI’s judgments

Discuss or debate how to address the problem of data bias.

‌ Choose one of the problematic cases of AI data bias and suggest specific

Choose one of the solutions to the problem to minimize bias through discussion or debate cases and suggest a activities. specific solution!

Solution

Activity Wrap up

You can identify the main causes of data bias arising in AI machine learning and describe various problematic cases related to it.

You can create a data set and AI training model to deal with the issue of data bias and develop an AI program using it.

You can propose specific solutions to minimize the data bias of AI through discussion or debate activities.

Problematic cases of AI data bias

Data causing socially discriminations due to 1 the bias in data selection Data of imbalanced amount or wrong 3 selection

Details

You can evaluate AI models trained on biased data and explain why the selection of key data attributes is important when training AI models.

Implicit bias in data that has been 2 accumulated throughout the human history

Obsolete or incomplete data

Details

‌ ‌ [D ] is the plural form of Datum, which means a material, resource, and an argument, and refers to a 1

‌ ‌ [D ] [B ] refers to the state in which human biases and errors are reflected in data 2

collection of resources or a chunk of information✽ for proof, judgment, and decision.

without modification with respect to the training process and results of AI models. This indicates that unfair

standards are infused when data is selected, collected, classified, and used, and as a result, AI prediction results

also have bias.

‌ ‌ [D ] [M ] refers to overall management activities such as defining, organizing, 3

supervising, and protecting data within a company or organization.

2. Connect the main causes of data bias in machine learning of AI and their corresponding explanations with a line.

Skewed sample

Tainted examples

Limited features

Sample size disparity

Existence of proxies

If the size of the training data for the minority group is much smaller than that of the training data for the majority group, the characteristics of the minority group cannot

be accurately reflected.

If the feature that can serve as a basis for the judgment is not included in the training data, the judgment is made based on other characteristics with low relevance.

As a result of learning biased data, AI learns such bias as well.

After the first training of the initial biased data, the bias is strengthened in the process of retraining on predicted results based on the first training.

Even after the characteristic causing the bias is removed, the bias still remains because the removed characteristic is derived from other characteristics during the

training process.

Data Management

You can explain the need to manage the data used by AI technology.

You can explain appropriate procedures and methods to use data in line with the purpose of data collection.

You can suggest specific measures to minimize data bias over the entire process of data collection and use.

Ⅷ

Taking Responsibility

You will be able to explain why we need to determine the appropriate actor to assume responsibility in any process of an AI system.

‌ You will be able to suggest concrete ways to clearly determine who should be held accountable in relation to the AI system.

You will define the actors to be held responsible across all processes from development to use of AI systems, understand the need to make such judgments, and put forward specific measures to clearly identify the actors responsible for the operation, use, misuse, etc. of AI systems.

‌With the development of AI technology, we are witnessing a wide range of problematic situations in relation to the operation, use, and misuse of AI systems. What are the specific measures to clarify who should be held responsible for the problem?

I am innocent!!!

This court decides that you are guilty.

This court decides that you are guilty. I am the one who should be blamed …

Judge, why do you judge innocent people as criminals?

The judgment was made by referring to the precedent data based on AI. Thus, the judgment is correct.

Actually, I am the one who is guilty.

No, that is not true!

How will you be held accountable if you made a wrong judgment due to incorrect data?

What is responsibility?

‌Examine the concept of responsibility, one of the 10 key requirements.

Responsibility is the obligations and burdens associated with involvement in something. In a problematic situation, a person who is responsible is obligated to take action to resolve the situation. From a legal perspective a responsible party can be understood to be subject to sanctions for illegal activity or inaction. Responsibility means the nature or attitude of Taking Responsibility in a problematic situation. This concept implies the value of a human being who responsibly fulfills his/her jobs or duties and has the ability to act and judge correctly.

✽

The relationship between AI and responsibility

The development of AI technology provides benefits of increasing people’s convenience and efficiency in decision-making, but can also cause physical and property damage. To minimize the negative impacts of AI technologies, actors responsible for developing and operating AI systems should behave responsibly.

AI technology influencing decision-making in healthcare

AI developers should make their utmost efforts to develop high-quality AI technology from the development stage to the result generation stage, and are responsible for faithfully following ethical procedures. AI service providers should assume shared responsibility for social and ethical problems arising from the dissemination and spread of AI technology, and are also obliged to take responsibility for malfunctions and accidents and to establish principles for compensation. AI users should fulfill their ethical responsibilities for the results of their use of AI, and should be entitled to reasonably hold AI developers and AI service providers accountable in the event of any safety accident due to AI’s own problems.

Sanction It refers to the restriction or prohibition of, the violation

✽

of certain rules or customs, or measures having such effects.

Think about the actors who are related to the

issue of responsibility for AI.

Discovery of a malfunction in AI system

AI service provider

AI training data

Health Data

Development of AI in compliance with ethical procedures

AI developer

Holding AI developers and AI service providers accountable in the event of an ethical violation

AI user

The actor who should take the responsibility: AI Developers - Service Providers - Users

Need to determine who should take responsibility

It is important to clearly determine who should be held accountable in order to secure effective remedy for damage. Therefore, we should understand the need to clarify the responsible actor in all processes involving AI systems, and have the ability to explain the need for and limitations in recording and tracking the processes of AI system operation and decision-making. In addition, we should discuss concrete ways to identify the record and influence of human intervention at each stage of decision-making by the AI system.

Prescription A Prescription A Prescription A

AI training data

What is the reason for the different

prescriptions?

Diagnosis of patient condition based on data

The need for human intervention in the decision-making process of AI systems

AI training data

Information feedback

Medical AI system upgrade

Case

Medical treatment and prescription by AI doctors - Who will be responsible for medical accidents?

◯◯ Hospital introduced ‘Watson,’ the first AI cancer treatment system in Korea in 2016. Based on data on historical patient cases and medical expert data, Watson recommended the optimal treatment method for cancer patients who are scheduled to receive anticancer treatment, which suggested the possibility of using AI as a means of assisting medical staff in their judgment.

Case

What if the following problem occurs? Since the introduction of the AI system, the consensus rate of diagnosis opinions between human medical staff and Watson has gradually increased, but cases where humans and AI diverge in judgment and prescription have continued to appear. Recently, it is also argued that the judgment ability of medical care systems such as Watson might exceed the human ability in certain areas. In this situation, concerns are growing that it is unclear who is responsible for medical accidents or unexpected side effects caused by humans’ excessive reliance on the judgment of the AI system in treating patients.

Briefly describe your opinion on who should take the responsibility.

Which diagnosis should I

My diagnosis is A. follow?

My diagnosis is B.

Algorithmic errors of robo-advisors in the finance sector, who should be responsible?

These days, AI robots are functioning as asset management advisors whose competency is comparable to that of humans in the financial sector. The robo-advisor engine is operated based on data accumulated by financial companies and their own algorithms, and proposes customized investment ideas or manages assets by itself.

What if the following problem occurs? The algorithm constituting a robo-advisor is like a black box whose operation process is not disclosed, so it is not easy to determine whether there is an error or at which point the error has occurred, and identify who is responsible even in the occurrence of any damage. This implies that even when users suffer damage due to errors and defects in the algorithm, there might be another problems such as difficulty in identifying the person to be held responsible or the absence of a monitoring system. In addition, since companies might attempt to escape from their responsibility by highlighting the black box problem of algorithms, some claim that a dedicated department or person in charge of complaints for damage should be designated.

Briefly describe your opinion on who should take the responsibility.

‌ The black box problem will be covered in Lesson 11, ‘Transparency’!

Who would take the responsibility?

Trolley Dilemma and Ethical Choices

Which one should we choose? Suppose you are the driver of a trolley car hurtling down the track at sixty miles an hour. Up ahead you see five workers standing on the track, tools in hand. You try to stop, but you can’t. <omitted> Suddenly, you notice a side track, off to the right. There is a worker on that track, too, but only one. You realize that you can turn the trolley car onto the side track, killing the one worker, but sparing the five. What should you do? Most people would say, “Turn! Tragic though it is to kill one innocent person, it’s even worse to kill five.” Sacrificing one life in order to save five does seem the right thing to do. Now consider another version of the trolley story. This time, you are not the driver but an onlooker, standing on a bridge overlooking the track. Down the track comes a trolley, and at the end of the track are five workers. Once again, the brakes don’t work. <omitted> Then you notice, standing next to you on the bridge, a very heavy man. You could push him off the bridge, onto the track, into the path of the oncoming trolley. He would die, but the five workers would be saved. (You consider jumping onto the track yourself, but realize you are too small to stop the trolley.) Would pushing the heavy man onto the track be the right thing to do? Most people would say, “Of course not. It would be terribly wrong to push the man onto the track.” <omitted> Why does the principle that seems right in the first case—sacrifice one life to save five—seem wrong in the second? If, as our reaction to the first case suggests, numbers count—if it is better to save five lives than one—then why shouldn’t we apply this principle in the second case, and push? It does seem cruel to push a man to his death, even for a good cause. But is it any less cruel to kill a man by crashing into him with a trolley car?

Let's think in light of the moral philosophy of Bentham and Kant.

Jeremy Bentham, An Introduction to the Principles of Morals and Legislation

Principle of utility principle of utility is meant that principle which approves or disapproves of every action whatsoever according to the tendency it appears to have to augment or diminish our happiness. <omitted> By utility is meant that property in any object, whereby it tends to produce benefit, advantage, pleasure, good, or happiness, or to prevent the happening of mischief, pain, evil, or unhappiness to the party whose interest is considered.

<Source> Michael Sandel, An excerpt from Justice

Immanuel Kant, The Groundwork of Metaphysics of Morals

Now I say that the human being and in general every rational being exists as an end in itself, not merely as a means to be used by this or that will at its discretion; instead he must in all his actions, whether directed to himself or also to other rational beings, always be regarded at the same time as an end.

책임성

Determine who should take the responsibility for AI systems

‌Give your opinion on who should be held accountable for AI systems in the following situation. Also

if the court rules that the driver should take responsibility, predict the future impact of this judgment

Fatal accident due to autopilot

A driver who was involved in a fatal accident while driving a car on autopilot, a self-driving assisting function of a car from Company T, has been charged with murder. This is the first time in Country A that the crime of murder has been applied to a normal driver who used the self-driving function. According to foreign press, the prosecutors of Country A indicted Mr. K on charges of accidental murder for crashing into an oncoming car while driving on autopilot, resulting in the death of two people. Mr. K had a criminal record of speeding at an intersection two years ago in December 2019 by driving on autopilot and ignoring traffic lights. Autopilot is a function that assists the driver in terms of accelerating, braking, and steering the car. Currently, the number of vehicles equipped with this function in Country A is about 765,000. Autopilot is a function that assists the driver, due to the misperception of some drivers who misunderstand it as a ‘fully autonomous driving function,’ there has been a string of related accidents. Currently, 26 autopilot-related accidents that took place after 2016 are being investigated by the Road Traffic Safety Bureau of Country A. These accidents were reported to have taken the lives of at least 11 people. An official from the Road Traffic Safety Bureau commented, “There are no cars available in the market yet that are equipped with full self- driving functions. All cars, with or without self-driving functions, must always be controlled by a human driver.”

Opinion on who should take the responsibility

Prediction of the influences on human society

Driver

Autopilot

<Source> The Chosun Ilbo, Sep. 01, 2019

Case 1

Developer

Service provider

User

Priorities of the responsible actors

Reason for the judgment on priorities

Set priorities in responsible actors for AI

‌Let’s find examples based on our own research. Based on

the contents of the research on the issue of responsibility

for AI, analyze opinions on responsibility from each actor’s

perspective, and set priorities among responsible actors.

Case 2

Developer

Service provider

User

Priorities of the responsible actors

Reason for the judgment on priorities

By examining a single problem situation from the perspective of multiple responsible actors, you

can broaden and deepen your thinking about the responsibility

for AI systems.

Discuss or debate from perspectives of several responsible actors for AI.

‌Select an issue related to the responsibility for AI, and have a discussion or debate on the issue

following the instructions.

Selection of a topic

Organization of groups and 2 assignment of roles

by holding a class meeting. Select a representative issue related to the responsibility for AI

service provider, and user).

responsible by each group (select one from an AI developer, Organize groups of 3 to 4 people, and select an actor to be held

Design of a discussion strategy and Do research on each selected actor, design discussion strategies through group discussions, and prepare related materials. 3 preparation of relevant materials

Initiation of discussions and 4 preparation of questions

Selection of a topic

Selection of a topic

Group composition

• Group name

• Members

• Role assignment

Wrap-up of the activity and provision 5 of feedback

feedback for the overall discussion or debate activity.

Organization of groups and assignment of roles

Actor to be held accountable

Have three groups participate in a discussion from the perspectives of an AI developer, a service provider, and a user

respectively, and ask for the remaining students to listen to the discussion as the panel and prepare discussion questions.

Make a final decision on which actor should take the responsibility for the problematic situation, and provide

Conclusion

• Judgment result

You can identify the responsible actors related to the AI system and prioritize such responsible actors.

Design of a discussion strategy and preparation of relevant materials

Claims and supporting facts

Preparation of discussion questions

Participant materials for discussion

• Actor to be held accountable

• Assertions and grounds

• Relevant questions

Wrap-up of the activity and provision of feedback

Feedback materials for discussion

• Analysis of reasons

• Prediction of the influences on future society

You understand that it is difficult to clearly determine which actor should be held accountable for an AI system, and analyze the relevant situation.

You can select issues related to the responsibility for AI and judge who should be responsible through discussions and debate activities.

Source

Experience ethical judgment using Moral Machine

Use Moral Machine and put yourself in the situation requiring ethical judgments on self-driving cars.

Moral Machine is a web platform developed by the University of Massachusetts (MIT), USA for the purpose of collecting social awareness of AI’s ethical decisions!

How to experience the Moral Machine

• ‌Ethical dilemmas that self-driving cars will face randomly appear on the screen. For example, a self-driving car without a vehicle occupant can recognize the presence of both pedestrians and vehicle occupants in the lane ahead and their approximate identities. However, the self-driving car senses that the brake is not working and under this circumstance, the car is given two options: either continue driving straight ahead and hit the pedestrian, or turn around and hit the pedestrian in the other lane. • ‌In some situations, one or more occupants are in the self-driving car. In this case, there may be an obstacle in one of the two lanes, affecting all occupants. In situations where one or two crosswalk signals exist, whether the pedestrian has violated the law can be considered as well.

• ‌The participant decides which choice the self-driving vehicle should make between the two given options. In each of the two choices, the affected character is indicated as a skull, medical cross, or question mark symbol representing death, injury, or uncertainty of outcome, respectively. • ‌The participant should choose an acceptable result between the two options in order to move on to the next situation. When you hover the mouse over a selected example screen, a highlight appears, and if you click that highlight, your selection will be submitted. You can see a summary of the results by pressing the button below each example screen, and the progress of the situation is also displayed numerically.

<Source> https://www.moralmachine.net/hl/kr, Awad, E., Dsouza, S., et al. The Moral Machine experiment. Nature 563, 59– 64(2018). https://doi.org/10.1038/s41586-018-0637-6

In this case, the self- driving car, whose brake What choice should a self-driving car make? moves straight ahead and suddenly stop working, in the front lane. crashes into a pedestrian

• 1 male athlete The result is … • 2 male managers • 1 male doctor • 1 male A pedestrian who will be in violation of the traffic affected is jaywalking signal.

Example situation of the Moral Machine experience

In this case, the self- driving car, whose brake changes its direction and suddenly stop working, in another lane. crashes into a pedestrian

• 1 female athlete The result is … • 1 female manager • 1 female be affected is crossing the The pedestrian who will crosswalk by obeying the traffic lights.

Analyze the results of ethical judgments in the Moral Machine.

‌Based on the results of the Moral Machine experience, verify the ethical judgment standards

and values of others and yourself, and think about why social consensus must be drawn in the

Category

The character who is saved the most

The character who is sacrificed the most

The importance of the number of victims

Preference for passenger protection

Preference for compliance

Preference for the avoidance of intervention

Preference for gender

Preference for species

Preference for age

Preference for physical condition

Preference for social values

process of developing an AI system.

Mark the results of Moral Machine experience in the table.

Result

Not important

Not important

Not important

Not important

Male

Humans

Children

Healthy

High

Very important

Very important

Very important

Very important

Female

Pet

Elderly people

Obese

Low

Write down your ethical judgment standards and values based on the analysis of the results of the Moral Machine experience.

‌Write down your ethical judgment standards and values based on the analysis of the results of the Moral Machine experience.

Category

Similarity

Difference

Your impression

Details

Read the following passage, think about why social consensus is required in developing AI systems, and predict the impact such social consensus will have on AI development, operation, and users.

The issue of ‘self-driving car ethics’: what are the answers of 2 million people?

The MIT Media Lab in the US collected ethical judgments and decisions from participants through the Moral Machine. The first element was, of course, that saving human lives should take precedence over saving animals. The second one was that a large number of people should be saved in preference to the few, and the third is that young people should be saved in preference to older people.

Researchers did not find any difference in moral judgment among participant groups classified based on demographic characteristics. However, they found larger clusters with somewhat different moral priorities from cultural and geographic perspectives. According to ethical and moral standards, the researchers were able to roughly divide the world into “Western,” “Eastern,” and “Southern.” Moreover, there were differences in ethical priorities among these three large clusters. For instance, countries classified as the ‘Southern’ cluster had a strong tendency to save younger people in preference to older people, and this made a sharp contrast to the ‘Eastern’ cluster, which was particularly respectful of the elderly. In theory, these priorities can be reflected in the software to be installed in self-driving cars. The question is whether it is desirable to reflect these differences in priorities when accepting new technologies. Surely, we cannot rule out the possibility that this experiment contains some biased results. Also, since the survey was conducted in the form of an online game, the opinions of young people could have been reflected more strongly than the those of the elderly. One of the most concerning scenario is that these results can affect autonomous decision-making principles of self-driving cars on the grounds that they are derived from a study conducted on a “global, large-scale” basis. There is another limitation that while ethical issues to be judged in real situations are very complex and subtle, these situational characteristics are not properly reflected in the study. For example, MIT researchers said that generally, ‘young people are saved in preference to old people,’ but the age boundary is very ambiguous. Furthermore, two or more criteria may overlap in some cases. Criticisms can be raised as to whether it is appropriate to prioritize simply by considering age when there are people crossing in compliance with the traffic signal and people jaywalking. <the rest omitted>

<Source> Science Times, Oct. 29, 2018

Make ethical guidelines on self-driving cars.

‌ Set up the guidelines for ethical design to ensure that self-driving cars based on AI systems can drive on roads in compliance with traffic laws without any problem. Also

Imagine va rious predict problems that may occur in driving automation situations where the created ethical dilemmas that self- driving cars might face, guidelines are observed, and suggest specific solutions to them.

and make guidelines for ethical design!

Who will take the

There is nothing to worry about thanks to data and algorithms.

AI, please drive the car for me!

What happened?

a guideline for I hope there is

Yes, definitely. Everyone

Is it a car accident responsibility? I will take There is no I am just a I just used the again? I think I am the responsible person to the police office with me. problem in provider of driving automation seeing too many self-driving car escape from the encourage mutual design! the service! function with my firm belief in AI.

Why don’t we I agree. That Why don’t we look for look for solutions sounds like a solutions instead of instead of denying responsibility and great idea. denying responsibility and fighting? fighting?

accidents.

Developer Service provider User

subjective Here thoughts. Would are our you like to hear opinions. our thoughts?

ethical design to

seems to

understanding.

responsibility.

Right

I think this contains quite a lot of your

design of a self-driving car No matter how perfect the

when human intervention is, there may be moments

is required for judgment.

Check out the characteristics of each level of Taking Responsibility driving automation on Page 44!

Predict and discuss four ethical dilemma situations self-driving cars might face.

Select one of the ethical dilemma situations predicted in 1 that you would like to deal with, and list the judgment criteria that may affect the decisions of self-driving cars.

Priority

Reason for selection

Prioritize the judgment criteria and write the reason.

Establish ethical design guidelines for self-driving cars based on the answers in 1- 3

AI developer

Service provider

User

After group presentations, compare, modify, and supplement the established ethical design guidelines for self-driving cars.

Predict additional problems that may arise in driving situations when the finalized ethical design guidelines for self-driving cars are applied, and suggest specific solutions.

Activity Wrap up

You can set up the ethical design guidelines for self-driving cars from the perspective of each responsible actor for the AI system.

system, and its users.

Details

You can experience judging on various ethical situations presented by the Moral Machine, and explain your own ethical judgment standards and values based on such experience.

You can explain why social consensus should be reached in developing a self-driving car system and how the social consensus will affect the development and operation of the

Rating

‌ ‌ refers to obligations or burdens to be taken or sanctions to be imposed as a result of being involved in something. 1

‌ [A ] [D ] should make their utmost efforts to develop high-quality AI technology from the 2

development stage to the result generation stage, and are responsible for faithfully following ethical procedures.

‌ [A ] [U ] should fulfill their ethical responsibilities for the results of their use of AI, and 3

should be entitled to reasonably hold other relevant actors accountable in the event of a safety accident due to AI’s

own problems.

‌ [A ] [S ] [P ] should assume shared responsibility for social and 4

ethical problems arising from the dissemination and spread of AI technology, and are also obliged to take

responsibility for malfunctions and accidents and to establish principles for compensation.

2. Read each of the following statements and mark ○ if correct and × if incorrect.

‌It is important to clearly determine who should be held accountable in order to secure effective remedy for damage. Therefore, we should understand the need to clarify the responsible actor in all processes

involving AI systems, and have the ability to explain the need for and limitations in recording and tracking

the processes of AI system operation and decision-making.

( )

‌Moral Machine refers to a web platform developed by the University of Massachusetts (MIT), USA for the purpose of collecting social awareness of AI’s ethical decisions.

( )

You can use problematic cases related to the responsibility for AI systems to seek solutions.

Responsibility

You can explain why we need to determine the appropriate actor to assume responsibility in any process of an AI system.

You can suggest concrete ways to clearly determine who should be held accountable in relation to the AI system.

Ⅸ

Ensuring Safety

‌You will be able to understand the risks associated with AI and investigate solutions by looking into cases related to the safety of AI systems.

‌You will be able to discuss the need for safety devices to address the problems can be caused by AI systems and find practical solutions.

‌You will be able to explore the ways to ensure the continuous cooperation among AI operators, developers, and users required to keep the performance of AI systems stable and consistent.

There is a problem in the AI system. I

need to examine it immediately!

You will understand that it is crucial to ensure safety in the overall processes for the development and use of AI systems, and seek various measures to ensure safety.

‌In contrast to general electronic products, AI systems might trigger great repercussions when they have errors in prediction and judgment. Therefore, these systems need a procedure for safe management and verification. What

Wow, is this the AI robot which protects us?

Look at this. It also has a school guidance function and a fever check function.

It would be very convenient!

It is trustworthy.

measures can be taken to enhance the safety?

In the following situation, what caused the safety issue?

A few days later

check today’s I am going to lunch menu!

This is amazing. Can I get answers to all my questions?

What happened, teacher? Phew. I Teacher!! What’s stopped it. Are you scared? wrong with the robot?

‌

‌ ‌Examine the concept of safety, one of the 10 key requirements.

What is the safety of AI systems?

AI systems should be designed and used with safety in mind. In this context AI should always operate in a way to ensure the treatment of all people with dignity. It should not act in such a way as to cause physical or emotional damage to anyone.

Because AI learns to operate based on data and not a set of programmatic rules-based instructions, it is possible for it to act in a way the developer does not expect. It is difficult to determine why some AI decisions were made.

For example, suppose a user travels the same road every day in a self-driving car. Input data to the AI system will vary each day based on such events as the weather and the movement of other drivers. It is impossible to predict, when designing the AI system, all the possible inputs that might occur and so the AI system’s response cannot be guaranteed. For this reason, safety may involve in this example, giving users the capability of controlling the car in the case of an obvious error.

In 2016, a 16-month-old baby was attacked by a security robot at a shopping center in the United States.

Beep!

‌

Cause A security service robot misperceives a 16-month- old infant as an enemy and attacks.

Cases of safety problem caused by recognition error of AI

In 2013, a car heading to a wedding hall in the Middle East was suddenly attacked by a drone, causing the deaths of 10 people.

Remove the risk!

Cause Accidents caused by error in judgment of AI systems

•‌Should ensure safety to prevent AI technology from negatively affecting humans. •‌Should comprehensively supervise the exceptional cases.

•‌Should carry out continuous quality control. •‌Should secure control devices for possible malfunctions and hazards.

Unlike general electronic products, AI systems can trigger great repercussions when they have errors in prediction and judgment. However, in practice, it is difficult to completely rule out the possibility of malfunction or accident. Therefore, we need a procedure to safely manage and inspect these systems.

Issues threatening the safety of AI systems

Since AI systems use computers and communication networks and have no environmental or time limitations, they can easily become targets of hacking. For example, self-driving cars can be affected by malicious traffic accidents or hacking attacks because they are operated on network and are controlled by trained data. In addition, AI systems are becoming more vulnerable to external attacks and attack speed is accelerating. For instance, spear phishing* selects an attack target from which it can earn a lot of value through an attack, and collects and analyzes information on the attack target to increase the attack success rate. Then, it begins attacking the target by impersonating as a person, company, or content that the target is interested in.

Efforts required to secure the safety of AI systems

In order to minimize the risk of AI malfunctioning or causing unexpected errors, both the system and the process of use should be inspected. That is, safety measures for the overall AI system and risks that may arise from human interaction (surveillance by AI, addiction or over reliance on the technology, etc.) should be examined and relevant risks mitigated prior to deployment of the system. The efforts of each actor, such as an AI developer, operator, and user, are required as well.

AI user

AI developer

AI operator

Efforts required to maintain the safety of AI systems

*Spear phishing In contrast to phishing, which steals personal information of an unspecified number of people, this refers to a phishing attack that aims to steal information from a specific person, and mainly uses targeted email attacks that appear to be from trusted senders.

•‌Should not arbitrarily manipulate AI technology and related products beyond the permitted range, or misuse them to infringe on the rights of others. •‌Should obtain information on the correct usage of AI services he/she uses and strengthen the

ability to utilize AI technology.

•‌Should identify risks in the distribution process and conduct thorough examinations in prior. •‌Should prepare safety verification and control measures.

•‌Should guarantee the freedom to choose for users.

Error in recognition of self-driving cars

IT media company T reported, “Automaker V faced an unexpected challenge while testing a self-driving car in Australia. The risk of an accident is increasing due to the self-driving car’s failure to recognize the kangaroo correctly.” A technical manager at Company V explained in an interview with Company A, “Kangaroos move by jumping. In this process, the sensor in self-driving cars often malfunction. The sensor is recognizing the object as being farther than the actual distance when it is floating in the air, and as closer when it landing on the ground.”

What was the problem?

Test results of self-driving cars verified that the system fails to perfectly detect kangaroos which move by jumping, unlike other animals, such as deer and elk, most of which move right on the ground.

Think about and briefly describe the cause of the recognition error in self-driving cars..

Robot that fell on an escalator and injured a person

<Source> Asia Economy Daily, Mar. 22, 2022

There was an accident in which a robot, which provides information at shopping malls, monitors people’s body temperature to prevent the spread of COVID-19, and entertained children by singing and dancing, took a wrong path and injured a person. At a shopping mall in Country C, a robot fell on an escalator on Christmas Day last year, and caused a personal injury. The exact reason why the robot moved in the wrong path has not been identified yet. There is also a dispute on whether or not there was external interference.

What was the problem?

A video of the AI robot falling off an escalator due to its malfunction went viral through social media. After the accident, the use of the robot was suspended, and the robot provider is said to be investigating the cause of the accident. The name of the provider has not been identified.

‌Think about and briefly describe whether users should be given the authority to control the function of AI systems

when there is a safety issue with them.

<Source> AI Times, Jan. 08, 2021

ase 3 C

Cars at greater risk of hacking as more electronic devices are equipped

‘Connected cars, also called “running smartphones,” are projected to become a new platform as they are combined with more information and communication technologies in the future. Accordingly, the cyber attacks targeting cars are also diversifying. A hacker can manipulate the system to lock or unlock the door, or turn the engine on or off. If a self-driving car is attacked while stationary, the damage is limited to physical theft, but a cyber attack on a driving car might cause greater loss of life. In particular, in the era of self-driving cars, where the system automatically handles steering, deceleration, and acceleration, the scale of potential damage is even greater.

What was the problem? In fact, a white hacker pointed out the possibility of such hacking. In May of this year, British cyber security company N discovered a security vulnerability in millions of digital locks where the Bluetooth Low Energy technology is deployed. The company succeeded in unlocking a 2021 Model Car Y from Company T, adding that Bluetooth Low Energy technology is not suitable for locking due to its own vulnerabilities.

Find the cause of the problem and briefly describe it.

<Source> Aju News, Jul. 04, 2022

suggestions for securing safety.

Case study

Case

White color is difficult to recognize!

Find possible cases of judgment error in self-driving car systems.

‌Investigate problematic cases caused by errors in

judgment in self-driving car systems, and write down

A case in which a self-driving car collided with a large trailer with white sides at a highway intersection.

You can find and paste pictures or photos with potential errors of judgment!

Make suggestions for Ensuring Safety.

Before starting the activity, refer to Page 44 for levels of driving automation and driver characteristics!

A case in which a self-driving car recognizes a person crossing a crosswalk with a green round bag on his back as a traffic light.

green traffic AI detects a

light. Let’s go!

Look into the principles of AI technology through an artificial neural network unplugged activity.

Experience the process of predicting and judging images by AI, and write reasons why safety

should be considered in developing AI systems.

In an artificial neural network, which refers to a machine learning technique that mimics human nerves, the step of sending input values is called the input layer, and the step of

receiving and outputting values is called the output layer. In addition, there is a hidden layer between the input layer and the output layer. In multiple hidden layers, the features

Unplugged activity

Necessary materials A pen and a sheet of paper

of the input image are extracted and inferred, and in the output layer, decisions on the images are made!

Input layer Hidden layer Output layer

…

Input data

Check the instructions for the unplugged activity.

Input data

Input layer

Hidden layer

Output data

Output data

Output layer

1 person to present pictures, 3 people for the input layer, Groups 1 and 2 (4 members Allocation of members each) to represent the hidden layer, 2 people to represent the output layer, and 1

person to judge the final prediction

How to play ① The member to present pictures selects one of the prepared pictures, and the remaining members except for the three representing the input layer stand facing back. ② The three representing the input layer look at the presented picture for 10 seconds, and after 10 seconds, turn the presented picture face down. ③ The three representing the input layer describe the characteristics of the photo in words on eight pieces of paper. ④ ‌Out of 8 pieces of paper written by 3 people for the input layer, randomly distribute 2 pieces of paper to each member of the group representing the hidden layer (4 people).

Input layer

Hidden layer

Output data

Output layer

⑤ ‌Hidden layer Group 1 selects 4 out of 8 sheets based on the most meaningful word (weight), and discards the remaining 4 sheets. ⑥ Hidden layer Group 1 randomly distributes 4 sheets of paper, one to each member of hidden layer Group 2 standing behind them. ⑦ Hidden layer Group 2 selects 2 out of 4 sheets based on the most meaningful word, and discards the remaining 2 sheets. ⑧ The people representing the output layer receive 2 sheets of paper, judge the image, and announce the result.

Check the difference between the final word and the picture.

If the word and the picture do not match, investigate at which point the error occurred and explain why safety should be factored into in developing AI systems.

Identify the causes of higher error rates of AI.

‌Investigate cases online where AI systems reveal errors in their predictions, and think about

methods to get rid of them.

Case study

Case study

Methods to get rid of errors

When the data does not contain enough information

Methods to get rid of errors

When unexpected data is entered

You can explain the factors that threaten the safety of AI.

You can explore and suggest ways to reduce errors in AI systems.

Discuss the need for safeguards and find practical solutions.

Read the following article on AI stop button development, and discuss and debate the need for

the development of a stop button.

‘Kill switch’ to stop ‘tyrant AI’ to be released

It is reported that Company G is conducting research on a ‘kill switch’ that forcibly shuts down the AI from the outside. In 2016, Company G developed ‘AlphaGo,’ an AI that defeated Se-dol Lee, Go player of 9 dan rank. This research on ‘forced shutdown’ is gaining great attention, as it implies the possibility that AI will become a major threat in the future. An official from Company G presented a paper titled “Actor that can be safely intervened in (by humans)” at an AI-related conference. Here, the actor means AI. Company M, a media specializing in information technology, introduced this paper and reported, “Researchers from Company G are developing a kill switch for AI.” A kill switch refers to a technology that remotely disables a smartphone or other device when it is lost, and is based on a concept similar to self-destruction. For AI with a “reinforcement learning” function that enables self-learning and discovery of the best answer, even developers do not fully understand its operation. This is the reason why AlphaGo, which was developed by researchers from Company G, whose Go skills are far inferior to Se-dol Lee, an experienced Go player, was able to defeat him.However, AI with these characteristics does not always work as intended by humans. A representative example is the case in which Company M’s chatbot learned and made racially discriminatory remarks. Bill Gates also warned of the dangers of super artificial intelligence by saying, “A very careful approach should be taken.” An AI with this dangerous potential risks disabling its stop button if a human discovers its problem and tries to stop it, deciding that human intervention is preventing it from coming up with an optimal answer. The main thesis of this paper is that it is possible to design a reinforcement learning AI that is incapable of resist human intervention through mathematical argument. However, the controversy is expected to continue. In a lecture, the CEO of electric car company T alleged that AI developed by tech giants could turn into a “tyrant” ruling over humans.

<Source> The Hankyoreh, Jun. 06, 2016

Instructions for activity

If you think the AI stop button is necessary, select ‘Support,’ and if you don’t, select ‘Oppose’ and write down the reason for your Select a position for the debate selection.

Group composition

Initiation of debate and wrap-up of the activity 4

Support

Oppose

Group composition

Group composition

Role assignment to each member

Organize groups for support and oppose positions of 3-4 people each.

Design of a discussion strategy and preparation of relevant materials 3

discussions, and prepare related materials.

against the need for a stop button.

Select a position for the debate

Why the stop button should be developed

Why the stop button does not need to be developed

Do research to find supporting materials for the relevant arguments, design discussion strategies through group

Have a debate among the groups from the perspectives of supporters and opponents, and summarize opinions for and

‘support’ and ‘oppose’ and write down the reason. Circle the position you chose between the

3 Design of a discussion strategy and preparation of relevant materials

4 Wrap-up and feedback after the debate

Summarize opinions

Support

during the debate, and provide feedback on the Summarize the opinions of both side presented

overall discussion and debate activities.

Summarize opinions

Oppose

Explore solutions to problems that can be caused by AI systems.

‌Based on the debate of the need for a stop button, write down solutions to the problems that AI

Write down the solutions.

systems might cause in the format of card news.

Card news refers to posts made to be viewed while swiping the screen composed of visual images and short texts one by one. This format delivers information by summarizing the

key points, making it quick to understand and easy to share.

Think of ideas for card news.

communicate in 9 pages or less and publicize it You can write briefly what you wish to

through the school website or social media.

Seek measures for cooperation among actors to ensure safety.

‌Read the following article on the need to establish the standards on reliability and safety for medical AI, and propose ways to cooperate to secure safety from the perspectives of AI developers, operators, and users.

Reliability and safety standards for medical AI required

The Korea Health Industry Development Institute recently released a report on the ‘reliability and safety of medical AI’ written by Hospital A. The report presented various concepts including explainable AI, robust AI, and fair AI to secure the reliability of medical AI. Furthermore, the report emphasized the need to make systematic legal systems and review relevant regulations on the grounds that safety is an issue that can have a significant impact on the scope of AI use. The ‘Reliability and Safety Guidelines for Medical AI Applications’ published in 2018 stated that principles such as protection of basic human rights, supervision of medical staff, robustness and safety, privacy and data management, transparency, diversity, fairness, prohibition of discrimination, medical development, responsibility, etc. must be observed, and the evaluation standards to ensure the fulfillment of these requirement should be established. According to the guidelines, AI must not infringe on human rights and autonomy, and the ability of medical staff to intervene and supervise the entire decisions made by AI should be guaranteed. In addition, medical AI should be safe and accurate. In particular, medical AI should be resistant to medical uncertainty, etc., should not be easily damaged by hostile external attacks, and should be reasonably reliable. To this end, medical AI should be equipped with minimum safeguards to defend itself from hostile or malicious attacks, and separate test results for such ability should be clearly set out. The European Union published the ‘D6.2 Guidelines on Regulating Robotics’ in June 2014. The main points of these guidelines are that to ensure the safety of AI, robots should be given the status of electronic humans and that AI should be developed with the aim of helping humans. In addition, the guidelines addressed the installation of a ‘kill switch’ that minimizes the potential social misuse of AI, including hacking, prevents any threat humans, and can immediately stop the AI system in case of an emergency. The organization also published ‘Ethical Guidelines for Trustworthy Artificial Intelligence’ in April 2019. The guidelines stipulate seven requirements for developing and using trustworthy AI, one of which is technological safety. In this regard, technical robustness is mentioned as the most important component. Technical robustness means developing AI systems to operate reliably by minimizing unintended risks and limiting such risks to an acceptable level. In order to secure the safety of AI, it should be examined whether the AI performs its tasks without harming any living creature or the environment, and unintended consequences and errors should be minimized. In addition, over various applications, a process to define and assess potential risks of AI systems should be formulated.

<Source> Medi Sobiza News, Nov. 5, 2005

Write down the reasons why the reliability and safety standards for medical AI are

required.

AI developer

You can propose the efforts required of actors and how they can cooperate to ensure the safety of AI systems through case studies.

Suggest the ways to cooperate to ensure safety.

AI operator

Details

You can consider and explain the need for a stop button for the safety of AI systems.

User

Rating

‌ ‌‌ [S ] is to minimize and avoid unintentional damage over the entire process of AI development and 1

utilization, and to prevent AI from undermining human dignity.

‌ ‌‌ [S ] [P ] selects an attack target from which it can earn a lot of value through an attack, 2

and collects and analyzes information on the attack target to increase the attack success rate. Then, it begins attacking

the target by impersonating as a person, company, or content that the target is interested in.

‌ ‌ [K ] [S ] refers to a technology that remotely disables a smartphone or other device when 3

it is lost, and is used to stop the AI if it malfunctions or operates undesirably.

2. Explain why malfunctions or unexpected errors occur in AI systems.

3. Mark with ○ for correct statement and for incorrect statements.

1 ‌Safety measures for the overall AI system and risks that may arise from human interaction should be assessed in prior, and the efforts should be made to mitigate these risks.

( )

2 Continuous cooperation among AI operators, developers, and users should be offered for the safety of AI systems.

( )

Self Checklist

Safety

You can explain the potential malfunctions and errors of AI systems and their causes.

You can recognize that AI systems are vulnerable to security issues and understand that AI systems should include safeguards to enhance their safety.

Ⅹ

Maintaining Transparency

‌You will be able to identify the impact that securing transparency in the process of development and use of AI will have on the formation of social trust.

‌You will be able to understand the risks people may face when using AI systems, and explore the scope and level of information users should be aware of to address them.

Check what users need to know!

You will think about what users should be aware of in developing and using AI technology, and find ways to ensure the transparency of AI.

‌AI is a technology that is still developing, and it has parts that are difficult for users to understand. What is the minimum information a user should be aware of to prevent damage arising from these characteristics? In addition, how can developers explain the technical characteristics of AI?

It says that a self- driving car that makes its own judgments and drives itself has been developed! This is

unbelievable!

Can AI decision making be made transparent?

Really? By the way, how does the car make such decisions? Is it true that the car distinguishes between a big animal and a child?

Unfavorable information may not be disclosed for the benefit of the company. Also, I wonder if we can understand the explanation, even if the company discloses it. the AI technology in this material are Oh, look at this. The explanations on very easy to understand.

Yes, it is. to develop explainable AI I think it will be important technology in the future.

Well, maybe we look at website of the car manufacturer and see how it works, don’t we?

‌ ‌Examine the concept of transparency.

AI and transparency

AI systems are expanding their territory, as they are used in job interviews which exert a major influence on job search, or in credit evaluations, etc. This means that recruitments or credit lines can change depending on the decisions provided by the AI system. However, due to the inherent complexity of AI technology, there are areas where it is difficult to identify the basis for the results, which undermines the reliability of AI systems.

Why transparency of AI is difficult to achieve

It is a technically challenging to track the operational process and principles of AI systems that are powered by machine learning algorithms. For example, a program that distinguishes whether an animal in a photo is an elephant or a rhino used to make judgments based on the rule pre-set by humans that “If it has a long nose, it is an elephant.” and “If a horn on its nose, it is a rhinoceros,” whenever a new photo was added. However, thanks to the advancement of machine learning, the system can now make its own rules to distinguish between elephants and rhinos, even if humans do not provide them in person. These rules are formulated by AI itself by taking a large volume of pictures of elephants or rhinos, and learning the correct answers for those pictures. The technology in which the system creates its own rules through observation and experience, that is, by learning the provided data and correct answers, is called machine learning.

Please explain the process that led to these

Ask the AI system!!

results.

Rule

programming Conventional

Data

Why was my loan

application rejected? declined because

Data

Correct answer

Correct answer

Comparison of traditional programming and machine learning

Recent advances in AI have occurred in a branch of machine learning employing what are called deep neural networks. These are programs designed to mimic some aspects of neuronal networks in human brain. Deep neural networks employ many thousands of neurons and require large computational systems to work. Because of this complexity, the rules an advanced deep neural network uses to make decisions are unknown even to the computer scientists that built it. This is called the "black box problem." Information goes in and results come out of a "black box" but the operations inside the box are unknown.

Bank

B

A

Bank

Are you dis- criminating

cation has been Your new appli- against me denied your loan

now?

you already have a large loan.

Rule

Machine learning

‌Black box problem

The review results are determined by the size of the loan or income, The AI algorithm and independent from other factors such as application based on your unstable education level, gen- der, or race. Check this data. income.

data for the Supporting

AI algorithm’s rejection

decision

When a customer who has applied for a loan asks the bank why the application was rejected, the relevant explanations should be provided transparently.

If the process a system uses to make decisions is ambiguous or incomprehensible it is difficult to determine what factors are responsible for the decision. Which actor should be held accountable in the case of a decision involving an accident or material damage if the cause of the problem cannot be identified? The lack of transparent operation undermines the reliability of the technology. 191

Efforts to make transparent AI

At the current level of technology, it can be considered that an inverse relationship exists between the accuracy and transparency of AI systems. Especially, in deep learning-based AI systems, the more complex the neural network structure becomes, the more difficult it is to secure transparency at the cost of higher accuracy. Despite these limitations, trustworthy AI is necessary for all humans living in the era of AI, and thus we should reinforce our efforts to improve and compensate for the opacity of the black box. In order to make transparent AI, AI system developers should develop explainable AI technology to boost the comprehensibility for users. In addition, AI system developers and service providers need to provide training or instructions on how to use AI, and share not only risk information, but also incidents of anomalies and side effects of the system.

Explainability

Developers or service providers should help those affected by algorithmic judgments understand the results by providing explanations of the judgments or disclosing the procedures for decision-making.

Measures to mitigate the black box problem of AI systems

In this era dominated by Artificial Intelligence (AI), it is imperative for citizens to demand enhanced transparency regarding AI products and services. Individuals should meticulously evaluate AI products or services, contemplating whether the developers or providers have accurately identified and incorporated users’ needs during the development process. Additionally, it is essential to assess the individual and societal impact resulting from the utilization of such products and services. If any uncertainties or queries arise concerning the use of these services, individuals, acting as consumers, have the right to proactively seek clear and transparent explanations from the developers or providers of the systems.

AI service developer and provider

Continuous verification should be implemented for the initial purpose of AI service development, the development modification process, etc.

Continuous monitoring is necessary to see whether user needs are identified and reflected in product development.

User

Mobility Transparency Committee announces taxi allocation algorithm verification results

The Mobility Transparency Committee (the Committee) visited K-Mobility without prior notice, checked the source code in the operating server where the service is running, and analyzed 1.7 billion call card sending history data. Through this investigation, the Committee verified whether the taxi allocation algorithm disclosed by the company was actually reflected in the system from four perspectives: allocation logic, source code, consistency between source code and server operation, and allocation logic operation status. As a result, the Committee confirmed that there is no discriminatory logic based on ‘taxi business type and passenger call distance,’ and the difference in allocation acceptance rates between affiliated drivers and general drivers stems from the difference in the system structures where affiliated drivers can see destination information and reject allocations, while general drivers receive automatic allocations without destination information.

ase 2 C

What are the difficulties in the transparency verification process? The Committee mentioned that it took six months to validate the source code of the K taxi allocation algorithm and added that a considerable amount of time was spent on negotiating with K-Mobility on whether or not to disclose the source code because the source code is a trade secret that contains the company’s core technology.

Think about what needs to be improved in the transparency verification process.

<Source> The Chosun Ilbo, Sep. 06, 2016

The number of studies on fairness and transparency of AI quadrupled in three years

The Institution of Human Centered AI (HAI) under Stanford University in US announced that the number of papers on AI ethics has quadrupled across the globe over the past three years since 2018, through the AI Index 2022. According to the report, the number of published papers on fairness and transparency of AI was aggregated as 302 worldwide in 2021 alone, which exceeds fourfold increase from 71 in 2018. In particular, the number of papers published by companies in 2021 amounted to 53, which increased by 71% from 31 in 2020.

What was the problem? The report analyzes that the reason for the increased interest in and research on fairness and transparency of AI is that as AI performance develops, the number of cases where sentences generated by AI include harmful expressions that demean gender, occupation, or race is rapidly increasing.

Write down the reason why the fairness and transparency is required in AI.

<Source> The Economist, May. 05, 2022

Learn about transparency through the case of US IT company I.

‌The following is an interview with Chief Data Officer of artificial intelligence strategy at Company I on how to build trusted AI technology. Read the article and find the point which is related to transparency.

technology.

US IT company I emphasizes the necessity of Trusted AI as solution to preventing and minimizing ethical problems for general companies which are examining plans to use AI

What is Trusted AI?

It means we are able to transparently verify AI models. Trust in AI will be defined by whether an AI model can monitor the bias embedded in itself, whether it can detect such risk, and whether it can be dynamically modified as needed. In order for companies to use Trusted AI, the AI must be transparent, explainable, and unbiased, must have a high understanding of business and ethics, for which, it should be based and its overall (data processing) process should be verifiable. In addition, AI systems on outstanding integration and well-managed data as well.

What kind of efforts is US IT company I making to implement trusted AI?

Expert Group on Artificial Intelligence (AI HLEG). We are providing the open source We have proposed a policy for responsible AI in the European Commission’s High Level community with tools and technologies, including ‘AI Fairness 360’ and ‘AI Explainability 360’ toolkits. At the annual meeting of the World Economic Forum (WEF) last year, we also urged the necessity of the precise regulation of AI. In cooperation with the Vatican, we helped plan the “Rome Call,” an international convention on AI ethics announced by the Vatican, and participated in the relevant convention. At this year’s Davos Forum, we decided to join the Global AI Action Alliance, which provides the guidelines for the responsibility policies, implementation activities, and partnerships required to ensure international cooperation in the design, development, and use of AI.

How can we recognize and eliminate discrimination and bias in AI?

and operating the model. It is important to have the ability to collect the model’s We should seek to define bias and then detect and reduce it in the process of building metadata and track the root cause of the bias. The lifecycle of a model should be documented to ensure continuous traceability even after the model changes and new versions emerge. If we can define the rules to verify the model in line with the company’s policies and standards, and ensure these rules are automatically enforced, Criteria for bias, fairness, and accuracy should be established as well when using the we can trust that the AI adheres to the standards of the company and the industry. model. Bias can be judged differently depending on the perspective of a country or industry, and can exist in many forms.

Do you think companies should disclose AI algorithms for trust and transparency?

Find a concept or statement that indicates transparency.

Please introduce a case where the measure for improvement of the AI model has been actually applied.

Last year, Company I worked with Seoul, Korea to carry out a policy development research project. They conducted research on actual conditions related to the daily lives of citizens which have been transformed due to the use of AI virtual assistants after the outbreak of COVID-19. In this research, along with an AI technology called W intent of questions and the basis of answers, was used. The researcher monitors the W Assistant, ‘W Open Scale,’ a technology that improves explainability to understand the provided, and recommends corrections. I think it is appropriate for AI like this to be Assistant’s conversation log to see what intention was applied and what answer was used in an automated, safe, and secure hybrid/multi-cloud environment.

I don’t think companies should be obliged to disclose the algorithm. What is important is that it is possible to understand how the results from the AI model held by the company have been produced, and explain it to individuals affected by the results. In the case of AI for the medical industry, it is necessary to be able to explain how the result was produced to doctors and patients, and in the case of AI used in the financial industry, to the transaction parties. This even applies to a case where the algorithm itself has not been disclosed. When simply checking the information which needs some explanations from the company, a ‘fact sheet’ tool included in the W OpenScale of the US IT company I can be useful. Like ‘nutritional facts labels’ indicated on the surface of food products, a fact sheet introduces three main components of artificial intelligence.

Is there any risk of confidentiality violation or know-how being exposed in an attempt to provide explainability?

Basically, even if our customer uses Company I’s tool to create an AI model, it does not guarantee that our company is completely free of bias, or give our company immunity from the problems caused by the AI model. However, we recommend our customers to leverage numerous tools and technical processes. We propose an approach that can minimize known biases and injustices when using the tool. We also discuss guidelines on what negative impact the use of this tool can have when creating AI models, what to do and what to avoid when bias and unfairness are found.

<Source> Aju Economics, Feb. 28, 2021

Make a fact sheet.

‌Choose one of the AI technologies or services you would use and make a fact sheet for the AI

product or services based on the terms and conditions.

A fact sheet is similar to the ‘nutritional facts labels’ that should be included on the label of a food product. This is a summary provided to enhance understanding of AI technology

AI use status

AI Product or service information

Collected data

Data management after the termination of the agreement

Division in charge of the product or service and its contact

Others

You can develop a fact sheet as an alternative in case it is difficult to disclose the algorithms of AI.

and services, and includes company overview, product or service information, key data, and other significant matters.

Fact sheet

You can find and explain various constituents of transparency of AI.

Who Wrote the Novel?

In August 2021, several media reported that an artificial intelligence (AI) novelist had published a full-length novel called <The World from Now On>. According to a report about a press conference on publishing, reporters seem to be dubious about whether this work was really written by AI. This is because the specific writing process or the information of the AI technology has not been revealed. On the cover, the names of both the AI novelist and the novel director are listed. On the book flap, their roles are divided into a ‘writer’ for the AI novelist and a ‘novel director’ for the writer. A note to the reader also says, “This book was written by an AI writer based on idea of the novel director,” clearly stating that AI wrote the novel. However, it is openly informed to the readers that the novel director wrote a part of the work himself and also decided on the title. However, only the name of the novelist is listed as “Author” on the copyright side that displays information about the book, such as the date of publication. It is a rather confusing situation where an AI novelist is a writer but not an author. Although it is difficult to judge whether an AI actually wrote a novel or not, and how valuable it is from the perspective of literature, it was clear that registering an AI as an author of a novel was not easy. The writer said that he conceived “the novel’s the theme, character and personality, point of view, and themes, not to mention the overall narrative structure” and the AI worked within the framework set by him. In the summer of 2016, news that a novel written by AI had passed the preliminary examination for a literary award in Japan made headlines. A lot of comments were made about AI, literature, and the future of humans, but the most impressive one was made by Professor Satoshi Sato, who participated in the development of the program. “If the part written by a computer is about 10 to 20%, this figure can be evaluated as the degree of contribution of the computer, but it does not matter if it is judged that it was written by a computer 100%. Also, since the programs were all developed by humans, I don’t think it would be wrong to say that the novel was written entirely by humans, not computers.” Over the past five years, sentence-writing programs have come a long way, but there has been little progress in our attitude toward AI as the actor of creation. Using AI in art creation does not make AI an artist. The effort to make a distinction between a novel written by AI and a novel written using AI is related to the difference in the will to take responsibility for the work rather than the difference in the aesthetic level of the work. AI may or may not be acknowledged as an actor of art creation, but at least someone should be able to put up their name, explain AI’s work, and be held accountable for it. This principle should apply not only to fiction, but to all areas where AI is used.

<Source> An excerpt from “Who wrote the novel?” under [Chi-hyung Jeon’s Somewhere Near the Science]

Recognize the necessity of transparency through a business case.

‌Read the following article about an US Automaker F, and analyze the costs and benefits of the

Company F’s choice. Also, write down your thoughts on the choice of Company F.

Company’s ethical dilemma

After the oil shock in the 1970s, the sales of small cars with good fuel economy significantly increased in the United States. In this situation, company F, one of the ‘Big 3’ in the automobile industry at the time, developed a small car named the Pinto and achieved great business success. However, when Pinto was hit from behind by another car, the vehicle’s fuel tank exploded and caught fire, often leading to the death of or severe burns to the driver or other vehicle occupants. The victims filed a claim for damages against Company F. After a tedious court battle, it was revealed by a whistleblower’s testimony that Company F had made an unethical decision. The company’s internal review and analysis document evidenced that although the company recognized that a fire could occur in the event of a vehicle collision due to a defect in the fuel tank, it had decided not to recall the cars based on the judgment that it was more economical to compensate individually when a lawsuit or legal claim was filed after an accident than the cost of calling and repairing cars that had already been sold was made public. As a result, company F lost the trial and was ordered to pay punitive damages of 125 million dollars (approx. 150 billion) in addition to 2.5 million dollars (approx. KRW 3 billion) in damages. Due to the so-called Pinto case, Company F’s corporate image ‘fell to the ground’ and it goes without saying that it lost consumer trust. We can conclude that the company, obsessed with short-term profits, forgot its corporate social responsibility of consumer protection, and as a result, it paid a severe price for ‘absence of business ethics.’ Establishing an ethical corporate culture including social responsibility will provide various benefits. The general public’s ethical expectations of a company should be met to increase its sales and production and enhance profits in the end. This concept is a modern advance from classical laissez-faire which is represented by Adam Smith who said in The Wealth of Nations, “If an individual or business pursues profit within the scope of not violating the law, the public welfare will naturally improve by the invisible hand.” Just as ethics that are valued and emphasized vary depending on times, business ethics can also evolve. These days when ethical management and social responsibility are emphasized, we should have wisdom to avoid a dilemma in which the pursuit of profit and business ethics collide.

Business ethics should come first!

Pursuit of interests should come first!

<Source> An excerpt from Maeil Business Newspaper, Sep. 5, 2022

1 Explore the concepts of costs and benefits of choice and analyze the costs and benefits of Company F’s choice.

Costs and benefits of choice

a choice means giving up something else. Therefore, in order to make a reasonable choice, we We face numerous choices in our daily lives. As the saying goes, there is no free lunch, making other words, we should thoroughly review the costs and benefits and make a choice where the should consider the gains and losses and select an alternative that gains more than loses. In Benefits refer to the gains or satisfactions we get from making a choice, which are expressed benefits outweigh the costs. bread and the non-monetary gains such as the psychological satisfaction a consumer gets from in monetary value. For example, both the financial gains a bakery owner earns from selling consuming bread are both benefits. Costs in benefit-cost analysis refer to opportunity costs. Opportunity cost is the benefit of the greatest value not actually chosen and given up, added by the amount paid for the actual choice. In other words, opportunity cost is the sum of what is given up and what is actually paid.

Choice 1

Choice 2

Instead of adding safety devices, pay $200,000 for Add $11 safety devices to repair 12.5 each fatality and $67,000 for each injured person million cars and trucks with defective in the event of an accident, assuming a fuel tank fuel tanks. failure would result in 180 deaths or injuries. In addition, expect 2,000 cars to be damaged due to the absence of safety devices, and set a repair cost to $700.

Choice 1

Choice 2

Cost-benefit analysis data of US Automaker F disclosed at the trial

Cost

‌12.5 million units × $11 = $137 million

Benefit

2 Summarize your thoughts on the choice of Company F.

How did US Automaker F deal with the fuel tank failure and what is the reason for the decision?

Suppose if you were an employee at Company F who discovered a defect in the car and you raised a complaint within the company

but it was not accepted. What should you do in this situation?

Suppose that you were the CEO of Company F and you receive a report of a complaint raised on the car within the company.

How would you solve this issue?

What is the attitude that is required for US Automaker F?

Set policy priorities for the development of transparent AI.

‌List the inconveniences you have to endure and the benefits you can gain from the measures to

Holding a public hearing to collect opinions before developing AI

Supporting the development of AI ethics policies

Other opinions

develop transparent AI and set their priorities.

Potential Benefits

You can verify the value of transparency through a cost-benefit analysis.

You can set policy priorities by analyzing the inconveniences you have to endure and the benefits you can gain to spread transparent AI.

Potential Inconvenience

Rank

1. Review key terms by finding terms that match the descriptions below and connecting them with a line.

A developer or service provider should increase their understanding of the results by providing an explanation of the judgment or the processes of decision-making to those affected by the algorithm judgment.

A method in which correct answers are output after rules are input and data is learned.

The more complex the structure of a neural network, the harder it is to get a clear picture of the rules or how it works due to the increase in computational workload.

2. Take initial letter quizzes and review key terms.

‌ ‌‌ [T ] means that the decision-making process in AI-based products or services. 1

‌ ‌ [M ] [L ] is a method in which a program learns given data and correct answers to create 2

its own rules.

‌ ‌‌ [D ] [L ]-based AI learns given data and creates rules on its own, but a huge amount of 3

complex calculations should be performed in this process.

Black box problem

Explainability

Conventional programming

You can increase transparency in the use of AI by preparing a fact sheet about AI products and services.

Transparency

You can identify the impact that ensuring transparency in the use of AI will have on the formation of social trust.

You can explain why transparency should be factored into AI technology in relation to the absence of explainable AI.

A

Abusive Language Detection Model …129 Adequacy evaluation … 34 Aggregation … 43 AI model training … … 105 AI speaker … … 31, 38, 104 AI-based recommendation platform… 53 AI-based senior care service … 101, 104 AI-based social credit system … 24 Artificial neural network … … 177 Attack by malicious programs … …… 67 Autopilot … … 156

B

Basic human right … 14 Bentham … … 155 Biased data … 138 Big brother … … 68 Big data … … 67 Biometric information … … 37 Black box … 154 Black box problem … 191

C

Carbon emission … … 105 Card news … … 183 Chatbot … … 36, 122 Company’s ethical dilemma … …… 198 Computer vision … … 94 Conditional driving automation … 44 Confirmation bias … … 53 Connected car … … 175 Conventional programming … …… 131 Costs and benefits … 199 Customized content… … 51 Customized service … … 33

D

Data… … 122 Data attribute … 138 Data bias … … 121, 122 Data management … 123 Data management process … 123 Data masking … … 43 Data misuse … … 66 Data quality management … … 124 Data reduction … … 43 Data set … … 138 Data suppression … … 43 Data vectorization … … 129 Datum … … 122 Deep learning … 192 Deepfake … … 16, 69 Defining problems … 92 De-identification … … 34 Design thinking … … 92 Difference … … 52 Discrimination … … 52 Driver assistance … … 44

E

Echo chamber effect … 16 Ego … … 32 Environmental impact assessment … 22 Ethical dilemma … 83 Ethical Guidelines on Self-Driving Vehicles … 164 Ethical vacuum … … 83 Ethics of responsibility … 103 Existence of proxies … … 123, 136 Explainability … … 192

F

Face recognition technology …15, 72, 125 Fact sheet … … 196 Failure of management … … 66 Filter bubble … … 53, 55 Filtering … … 131 Flaws in AI … … 67 Follow-up management … … 34 Fourth Industrial Revolution … … 52 Freedom of expression … 35 Full driving automation … … 44

G

GPT-2 … … 131 Guidelines for human rights … … 17

H

Hans Jonas … 103 Hidden layer … … 177 High driving automation … … 44 Human rights … … 14 Human rights impact assessment system … … 22

I

Ideate … … 92 Identifiable information … … 34 Information … 122 Input layer … … 177

J

Jamboard … … 94 Justifiable reason … … 73

K

Kant … … 155 Kill switch … … 180

L

Language model … … 129 Limited features … 123, 136

M

Machine learning … … 86, 190, 191 Mainstream culture … … 53 Mass media … 53 Misuse by governments … … 67 Mock trial … 74 Moral Machine … … 160

N

Natural rights … … 14 No driving automation … 44 No violation … … 66 Non-identifiable information … …… 34

O

Odysseus … 40 Online behavioral information … … 32 Operational definition … … 56 Opportunity cost … … 199 Output layer … … 177

P

Partial driving automation … … 44

Personal information … … 32 Population … … 122 Pothole … 81 Prepare follow-up measures … …… 131 Privacy … … 32 Privacy protection … 32 Privacy violation … 68 Prototype … 92 Proxy … 123 Pseudonymization … 43, 127, 128 Public nature … … 82

R

Ransomware … … 67 Recommendation on the Ethics of Artificial Intelligence … … 61 Respecting diversity … … 52 Responsibility … 152 RFID … … 67 Right to be forgotten … … 13 Right to know … 35 Right to self-determination of personal information … 35 Robo-advisor … … 154 Ron’s Gone Wrong … 92

S

Safeguarding human rights … 14 Safety … … 172 Sample … 123 Sample size disparity … … 123, 136 Sanction … … 152 Self-checklist for AI use … … 58 Self-driving car … … 44, 160 Sensitive information … … 33 Sign language … 85 Siren … 40 Skewed sample … … 123, 136

Smart city … … 19, 81 Social isolation … … 101 Solidarity … 102 Spear phishing … … 173 Stephen Hawking … … 116 Stop button … 180 Subculture … … 53 Surveillance society … … 67 Sustainable Development Goals (SDGS) 103, 106

T

Tainted examples … … 123, 136 Teachable Machine … … 112 Test … … 92 Thomas Kuhn … 82 Transparency … … 190 Trolley dilemma … 155

U

Universal Declaration of Human Rights … 14 Universal human rights … … 32

V

Vector … … 129 Voice-enabled traffic lights … … 84

W

Watson … 82, 154

Hans Jonas: https://kiss7.tistory.com/2671

• Page 10

Global Village: Shutterstock

• Page 14

Face recognition technology: Shutterstock

• Page 15

Echo chamber effect: Shutterstock

• Page 17

Case 1. Controversy over patented AI face recognition

• Page 18

technology: AI Times. 2022. 02. 15.(http://www.aitimes. com/ news/articleView.html?idxno=142912) Photo of Case 1. Face recognition: Shutterstock

• Page 18

Case 2. Controversy over human rights violations in

• Page 19

smart cities using AI systems: National Human Rights Commission of the Republic of Korea, ‘Study on Human Rights Guidelines in the Development and Use of Artificial Intelligence (Final Report)’ Photo of Case 2. Smart city: Shutterstock

• Page 19

Arguments for and against the AI-based social credit

• Page 24

system: Seung-eun Lee, ‘Comparison of Korean and Chinese University Students’ Perceptions of China’s Social Credit System’ (April 2019) Photo in Arguments for and against the AI-based social

• Page 24

credit system: Shutterstock AI systems helping the defense industry: The Chosun Ilbo.

• Page 27

2022. 08. 22.(https://www. chosun.com/economy/tech_ it/2022/08/22/ FFKWKDEATNDB5NAU6UUFIN6WKA/) The Guidelines on De-identification of personal

• Page 34

information: The Ministry of Public Administration and Security. 2016 Case 1. Photo of a chatbot that mimics a conversation

• Page 36

with a human: Shutterstock Case 2. Efficiency and risk of the use of biometric

• Page 37

database: The Chosun Ilbo. 2022. 06. 23.(https:// www. chosun.com/international/china/2022/06/23/ GB65FNBXKRANDPVDWETXGOTN7Q/) Photo in Case 2. Efficiency and risk of the use of

• Page 37

biometric database: Shutterstock JohnWilliamWaterhouse,Odysseusand

• Page 40

the Sirens, 1891 (https://ko.m.wikipedia.org/ wiki/%ED%8C%8C%EC%9D%BC:John_William_ Waterhouse_-_Ulysses_and_the_Sirens_(1891).jpg) • Page 45 Personal information protection, a prerequisite for vitalizing the self-driving car industry: The Law Times. 2017. 11. 16.(https://m.lawtimes.co.kr/ Content/ Article?serial=122840) Filter bubble: Shutterstock • Page 53 Case 1. Does AI prefer lighter skin?: The Hankyoreh. 2016. • Page 54 09. 14.(https://www.hani.co.kr/arti/ PRINT/760757.html) Case 2. Is it possible to guess the facial feature just by • Page 54 hearing the voice?: Ai times. 2022. 04. 28.(http://www. aitimes.com/ news/articleView.html?idxno=144324) Photo of <Ron’s Gone Wrong> (https://search.naver. • Page 55 com/search. naver?where=nexearch&sm=tab_ etc&mra=bkEw&pk id=68&os=24697533&qvt=0&qu ery=%EA%B3%A0%EC%9E%A5%EB%82%9C%20

• Page 61

• Page 61

• Page 67

• Page 69

• Page 69

Shutterstock

• Page 70

• Page 74

• Page 84

• Page 84

• Page 85

• Page 86

php?idx=13461)

• Page 92

Shutterstock

%EB%A1%A0%20%ED%8F%AC%ED%86%A0) UNESCO ‘Recommendation on the Ethics of Artificial Intelligence’: Korean National Commission for UNESCO Artificial Intelligence (AI) Ethics and Law - (I) Study on issues of AI ethics and governance Photo of UNESCO ‘Recommendation on the Ethics of Artificial Intelligence’ Shutterstock Amount of data created on the internet in a minute as of 2021: Visual Capitalist Case 1. Deepfake videos used in war: Herald Business. 2022. 03. 17.(http://news.heraldcorp.com/view. php?ud=20220317000561) Photo of Case 1. Deepfake videos used in war

AI popularization: The Fourth Industrial Revolution Committee, Survey on the use and perception of AI for its popularization, 2021. 06. Responses of each field in the United States to face recognition technology after the incident: Social Controversy and Implications of the Introduction of Face Recognition Technology: Focusing on the US Cases (ETRI Technical Policy Brief, December 2020) Case 1. Voice-enabled traffic lights with IOT technology: Anjun Journal 2022. 04. 11.(https://www.anjunj.com/ news/ articleView.html?idxno=32747) Case 1. What could be the problem?: The Kookje Daily News. 2021. 04. 19.(http://www.kookje. co.kr/ news2011/asp/newsbody.asp?code=0300&k ey=20210420.33003005763) Case 2. Why do we need AI for translating a sign language: Gonggam. 2022. 01. 24.(https://gonggam.korea. kr/ newsContentView.es?mid=a10221000000&section_ id=NCCD_POLICY&content=NC002&code_ cd=&news_ id=EBC6D4014C254203E0540021F662 AC5F) Deployment of machine learning technology to analyze the size of the economy in impoverished areas: Dongascience 2016 08. 2016 (dongascience.com/news.

Process of developing an electric wheelchair: Artificial Intelligence Times 2018. 12. 04.(https://www.aitimes.kr/ news/articleView. html?idxno=12974) • Page 104 Case 1. Offering of AI-based senior care service: Smart City Korea. 2022. 04. 06.(https://smartcity. go.kr /2022/04/06/%EB%82%98%EC%A3%BC% EC% 8B%9C-%EC%9D%B8%EA%B3%B5%EC%A7% 80%EB%8A%A5ai-%EA%B8%B0%EB%B0%98- %EC%96%B4%EB%A5%B4%EC%8B%A0-%EB%8F%8C%EB%B4%84-%EC%84%9C%EB%B9%84%EC%8A%A4- %EC%B6%94%EC%A7%84/) • Page 104 Photo of Case 1. AI-based senior care service

• Page 105 Case 2. The problem of rapid growth in carbon emissions due to training of AI models: IT WORLD, Jun. 12, 2019 (https://www. itworld.co.kr/news/124194#csidx6b67a799 b62078f 90b4b4d29da74572) • Page 106 Sustainable Development Goals: https://www.unesco. or.kr/data/ unesco_news/view/764/1094/page/0? • Page 107 Jennie, a companion dog robot: AI times. 2021. 05. 26.(http://www.aitimes.com/news/articleView. html?idxno=138606) • Page 110 PET bottle separation policy: Money Today. 2022. 04. 25.(https://news.mt.co.kr/mtview. php?no=2022042521375314456) • Page 117 Dr. Stephen Hawking: Dongascience. 2018. 03. 15.(https:// www.dongascience.com/news. php?idx=21755) • Page 122 Cases of AI’s bias: Dtoday. 2020. 6. 8.(http://www.dtoday. co.kr/news/articleView. html?idxno=366646) • Pages 123,136 The main causes of data bias in machine learning: Solon Barocas & Andrew D. Selbst. Big Data’s Disparate Impact(2016). Neural Information Processing Systems. • Page 123 Details of data management process: http://www.softline. biz/2018/sub02_01.html • Page 125 Case 1. Fine for misuse of personal information and data: AI times. 2022. 05. 24.(http://www.aitimes.com/news/ articleView. html?idxno=144774) • Page 125 Photo of Case 1. Fine for misuse of personal information and data: Shutterstock • Page 125 Case 2. AI bias: AI times. 2020. 06. 25.(http://www.aitimes. com/news/articleView. html?idxno=129812) • Page 125 Photo of Case 2. AI bias: Shutterstock • Page 126 Photo of AI chatbots: Shutterstock • Page 127 Privacy Policy for AI Chatbot Development: ScatterLab (https://team.luda.ai/ai-ethics) • Page 132 Collection,use and restriction of medical data: Bio Times. 2022. 05. 23.(http://www.biotimes.co.kr/news/articleView. html?idxno=7882) • Page 134 The need for big data-based intelligent education service and prospects for its introduction: Happy Education from the Ministry of Education(https://happyedu.moe.go.kr/ happy/bbs/selectHappyArticle.do?bbsId=BBSMSTR_ 000000005152&nttId=9273) • Page 152 AI technology influencing decision-making in healthcare: Shutterstock • Page 155 Trolley dilemma and ethical choices: Michael Sandel, <Justice> • Page 156 Fatal accident due to autopilot: The Chosun Ilbo. 2022. 01. 19.(https://www.chosun. com/international/ us/2022/01/19/ CFNJOCCEZ5H47ERV7G2N57ZN4A/) • Page 160 How to experience the Moral Machine: https://www. moralmachine. net/hl/kr, Awad, E., Dsouza, S., et al. The Moral Machine experiment. Nature 563, 59–64(2018). https://doi.org/10.1038/s41586-018-0637-6 • Page 163 “The issue of ‘self-driving car ethics’, what are the

Shutterstock

• Page 190 Photo of a black box: Shutterstock

campaign=biz)

answers of 2 million people?” Science Times, Oct. 29, 2018 (https://www.sciencetimes.co.kr/ news/ self-driving car ethics: what are the answers of 2 million people) • Page 163 Photo of “The issue of ‘self-driving car ethics’, what are the answers of 2 million people?” (https:// www. moralmachidne.net/hl/kr) • Page 174 Case 1. Error in recognition of self-driving cars: Asia Economy Daily 2022. 03. 2022.(https://zdnet.co.kr/ view/?no=2016-****-092842) • Page 174 Case 2. Robot that fell on an escalator and injured a person: AI times. 2021. 01. 08.(http://www.aitimes.com/ news/articleView. html?idxno=135396) • Page 175 Case 3. Cars at greater risk of hacking as more electronic devices are equipped: Aju News. 2022. 07. 04.(www. ajunews. com/view/20220703080835330) • Page 175 Photo of Case 3. Cars at greater risk of hacking as more electronic devices are equipped: Shutterstock • Page 180 ‘Kill switch’ to stop ‘tyrant AI’ to be released: The Hankyoreh. 2016. 06. 06.(https://www.hani.co.kr/arti/ science/ science_general/746899.html) • Page 180 Photo of ‘Kill switch’ to stop ‘tyrant AI’ to be released

• Page 184 Reliability and safety standards for medical AI required: Medi Sobiza News. 2021. 11. 05.(www.medisobizanews. com/news/articleView.html?idxno=82683) • Page 184 Photo of Reliability and safety standards for medical AI required: Shutterstock

• Page 193 Case 1. Mobility Transparency Committee announces taxi allocation algorithm verification results: The Chosun Ilbo. 2022. 09. 06.(https://biz. chosun.com/it-science/ ict/2022/09/06/4VXDL3UN KRFCJHCP3RVZSWY3SI/?utm_ source=naver&utm_medium=original&utm_

• Page 193 Case 2. The number of studies on fairness and transparency of AI quadrupled in three years: The Economist. 2022. 05. 14.(https:// economist.co.kr/2022/05/14/policy/ checkReport/20220514160010614.html) • Page 194 The article on the interview with Chief Data Officer of artificial intelligence: Aju News. 2021. 02. 28.(https://www. ajunews.com/ view/20210228100228237) • Page 197 Who wrote the novel?: https://www.hani.co.kr/arti/ opinion/column/1012498.html • Page 198 Business case in Recognize the necessity of transparency through a business case: Maeil Business Newspaper. 2022. 09. 05.(https://www.mk.co.kr/ news/business/ view/2022/09/783723/)

Authors

External Review Team

Dr. Sungae Kim (Duksung Women’s University) Dr. Soohwan Kim (Chongshin University) Dr. HanSung Kim (The Cyber University of Korea) Dr. Hyo-eun Kim (Hanbat National University) Dr. Sunyong Byun (Seoul National University of Education) Dr. Cheongho Lee (Sangmyung University) Dr. Adam Lewandowski (Asia Pacific International School) Emily Sonneland (Asia Pacific International School) Mary Kester (Asia Pacific International School)

Planning & General Supervision

Chapter V, VI, IX | Young-jun Kim (Sung Nam High School) Chapter I, II | Bang Sig Kang (Dongbuk High School) Chapter III, IV, X | Seo-hyun Lee (Soongeui Girls’ High School) Chapter VII, VIII | Seong-chae Choi (Changduk Girls’ High School)

Editor in Chief | Dr. Rosa (Hyun Kyong) Lee (KISDI, Department of Intelligent Information Society Policy Research) Project Coordination | Jihye Kim (KISDI, Department of Intelligent Information Society Policy Research)

Editing Team

Production Manager | Kyung Jin Cho (Cmass Communication) Assistant Editor | So Min Gwon (Cmass Communication)

Design

Cover Design: Unikkodesign Book Contents Design: Ki Bok Lee (Cmass), Mi Ra Lee (Cmass), Yeo Bi Lee (Cmass) Illustration: Yun Jai Kim (Cmass), Nam Young Ryu (Cmass), Hee Kyung Wuy (Cmass)

Artificial Intelligence Ethics (for High school students)

Creative Commons License (CC) 2024 Published by Korea Information Society Development Institute (KISDI) 18, Jeongtong-ro, Deoksan-eup, Jincheon-gun, Chungcheongbuk-do, 27872 Rep. of KOREA

ISBN (E-book) 979-11-7000-372-4 (95300)

** This publication has been supported by the Republic of Korea Ministry of Science and ICT