Middle school | For students

Artificial Intelligence Ethics

Introduction

Artificial intelligence (AI) ethics is evolving in line with the development of AI. AI can have a positive and negative influence, depending on how we develop and use it. From now on, we need to think ethically based on the technical principles of AI. This is because the many social problems in the age of AI will differ from the ethical problems we are facing in society today.

Thus, the demand for AI ethics education is gradually increasing. This change stems from the increase in the use of AI technology in our lives and the social and economic benefits and opportunities that are a direct result of AI. Learning AI ethics means forming a social consensus on how to use AI in a beneficial way 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. Although AI is developed for people’s benefit, as with all other technologies, adverse effects may result from the development and use of AI. In contrast to automated systems that only carry out repeated tasks, AI has infinite possibilities to judge and act in consideration of its surroundings and beyond the level of human intelligence. As AI generates both high expectations and fear, AI ethics has become more important than ever before.

Although it was initially believed that ethics should be programmed into AI, there has been a shift in understanding that people should be equipped with AI ethics themselves since humans are the subjects of creation and use of AI. For now, students are mainly playing a role as users, but they can grow to become AI developers or product managers of the future. Therefore, it is crucial to educate students going forward to help them think about AI ethics and put them into practice. Students will be required to observe the overall process of developing and using AI in the future.

In light of current education on the use of AI where ethics is treated separately, the aim of this book is to provide an integrated education. This book focused on helping students develop their own thoughts and judgment on potential ethical issues throughout the entire process of developing and using AI, compare and analyze their thoughts with those of their friends, and to establish an informed perspective.

This book will guide 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. Above all, this book explains the 10 key requirements of 「Artificial Intelligence Ethical Standards」 which have been compiled into one chapter, so that middle school students can reflect on these values through hands-on activities, with most topics composed of 4 sessions. At the same time, this book is organized by modules and sessions to make sure the necessary sessions can be individually separated for training on specific situations. The characteristics of each session are as follows.

• Session 1 Present problematic situations in relation to the 10 key requirements, and enable students to understand the

relevant key requirements and the roles of AI developers and AI users in the AI program development process.

• Session 2 Develop an AI program in line with AI ethics.

• Session 3 Identify possible ethical issues we faced in the use of the AI program developed in Session 2, and in the process of

seeking measures to solve such issues, analyze the differences between the perspectives of AI developers and AI

• Session 4 Perform activities to put AI ethics into practice.

users, and recognize the need to reach an agreement to narrow the gap in those perspectives.

However, since it is inevitable to present negative examples to teach AI ethics, there are concerns that students will come to have a negative perception of AI rather than learning the desirable directions for the development of AI. Nonetheless, we believe that the knowledge of the negative aspects of AI, which must not be neglected in the process of collecting and learning data, will contribute to a more realistic AI ethics education.

We hope that students will realize the significance of an ethical attitude and be able to behave in an ethical manner through this textbook.

Chief Writer Jeong-won Choi

Manage Personal Data for AI Development

• You will learn the process and method of collecting personal information for problem solving using AI models and collect and use necessary information.

Objectives Learning

What Is the Truth Behind the Face Recognition Powered by AI?

AI Story

Company A developed an app that uses facial recognition technology to identify the person shown in an image or video. This app is expected to provide the ability to monitor criminals or people on blacklists when used by the police, or when used by regular citizens to avoid the awkwardness that comes from forgetting the name of an old friend. However, civic groups and research institutes in the US urged the Company's CEO to suspend this face recognition service. This is because the app is likely to target people of color for spying or privacy violations. In fact, this face recognition technology looked at pictures of 80 state legislators in the U.S. and predicted that 26 of them were criminals. In addition, it has been also reported that many people could be chosen as suspects according to some allegations that the app is biased against women or people of color.

Questions for Critical Thinking

If AI benefits our society, should we put up with a few minor cases of privacy violation caused by AI? What if you are the victim of such violation?

Learning Process 2

STEP

STEP

STEP prevent any violation Protect privacy and

STEP Perspective of the developer who

STEP

user who provides Perspective of the

Understand the issue

Key requirements of AI ethics

collects personal information

(understand personal information)

personal information

Develop an AI Program

Program the “Make Friends Fast AI App” to ensure it functions correctly.

Developer

Program the App

Program the “Make Friends Fast AI App” to work as you planned.

You can use the command blocks to control the AI model by importing them from the 'AI' command block packs as follows

Structure &

Characteristics

What Problems Should Be Addressed?

Find out what problems we need to address in this lesson and how to solve them.

Protecting Privacy

Discover Problems in Our Daily Lives

After entering middle school, some students find themselves in a situation where they do not know most of their classmates at their new school because they have very few friends who have graduated from the same elementary school. Is there a good way to quickly become familiar with new friends and build close relationships with them in a new semester? If there is an app that collects students' face images and provides information on them to help users quickly learn new friends' faces and make friends with them, it can help students enjoy a better school life. If you develop a “Make Friends Fast AI App” where AI learns your classmates' faces and tells you who the other person is and their characteristics, what might be some issues that you should take into account?

name of that What is the

guy?

He is Sung- sik Yang in first grade, Class7.

hang out together Hi, Sung-sik! Let's

from now on.

Sounds great!

Write down what kind of issues we need to consider.

Think about and write down possible solutions.

Program the app to identify the face of a classmate shown in front of the camera and display the student's name and characteristics.

Use the completed app.

Show your in front of the friend's face

camera.

Understand Key Requirements of AI Ethics

Explore the given case and

learn the concepts of key

requirements of AI ethics.

Develop a Program to Understand AI Ethics

Experience the process of

developing the AI program needed

to solve problems.

Consider the Developer's Perspective More Deeply

Protecting Privacy

Make a plan for developing the “Make Friends Fast AI App”.

Developer

Implement an Additional Feature Using AI-based Face Recognition

Activity

It is expected that another app that offers more benefits can be developed using the face recognition function of the “Make Friends Fast AI App”. Consider adding additional features.

If we are allowed to use the feature for recognizing students’ faces, what kind of app can we develop?

In the consent form for the collection and use of personal information and the provision of third parties provided earlier, matters regarding additional features were not specified. Is it right to develop extra features? Since the relevant information has already been collected, can we use the information as long as we use it conscientiously and correctly?

Activities to Protect Privacy and Prevent Any Violation

• You will explore how to share your views on the balance between using AI for convenience and ensuring privacy protection.

Objectives Learning

Usage of AI vs. Privacy Protection, the Endless Controversy Is Still Ongoing

AI Story

Some regions of a country are operating a system in which financial information, compliance behavior, and online behavior data are used to record an individual's social credit, and the individual or related company is punished or rewarded based on the credit result. The right to purchase airline tickets or high-speed rail tickets, or to purchase accommodations or to dine out while traveling is also determined on the basis of social credit. However, some experts are concerned that this social scoring system will be used to spy on individuals, and violate their privacy and freedom.

Questions for Critical Thinking

Social development using AI or protection of privacy, which should be prioritized? What should be done to strike a balance?

Learning Process 2

STEP

STEP

Make traffic lights to protect privacy

Have a debate

Protecting Privacy

Developer

Address Unexpected Problems of the “Make Friends Fast App”

Activity

You are told that the “Make Friends Fast App” which has been developed to help users build friendship started to be used outside schools and unintentionally obtained personal information such as friends’ addresses and what they did with whom. Think about the following case.

the app to find out I am going to run I don't feel so

good.

who he is!

Hey, I know that yawn!

I feel sleepy!

He is Sung-sik Yang from Class 1. He likes pranks.

I didn't want anyone to recognize me I used the because I came out "Make Friends Fast App" and without washing

it introduced you to me.

my face.

You might learn information that people using the app have not consented to disclose information to others. Write down how you would feel as a developer if your app is used in a way that you did not intend for it to be.

Would it be better to destroy the app in the event of the problem as described in Question 1? Or should you correct the problem so that people can continue to enjoy the app’s benefits? Write down your opinion and the reason why.

Make Traffic Lights to Protect Privacy

Participate in activities where you can put into practice the ethics (requirements) you learned earlier.

Make Traffic Lights to Protect Privacy

Draw traffic lights and then write down your action plan for privacy protection in line with the meaning of each light.

Examples

Stop! Do not think “I don't care, because it is not my personal information.”

Go! Let’s consider the risk of privacy violation in advance!

Go! Let's share the risks of privacy violation with many people and prevent them in advance!

Go! Let's organize a system for quick response in case of privacy violation!

learned.

<Red light>

<Yellow light>

<Green light>

Never think “I don't It's risky to think Let’s consider the care, because it is "Let's collect risk of privacy not my personal personal informa- violation in information.” tion first and then advance! think about how to respond!"

Think from the Perspectives of the Developer and the User

Learn the AI ethics that need to be

considered from the perspectives

of the developer and the user.

Perform Activities to Put into Practice the Key Requirements of AI Ethics.

Wrap up the lesson by having

discussions on the key

requirements of AI ethics you have

2 that Recommends Clothes

Table of Contents

Why Do We Need AI Ethics?

Do We Really Need to Learn AI Ethics? 1 Why Should We Learn AI Ethics?

How Much Does AI Know about Your Behaviors?

Manage Personal Data for AI Development 1 Develop a “Make Friends Fast AI App” to Understand

2 the Protection of Privacy and Prevention of Any Violation 40

What Do AI Developers and Users Think?

47 3 Activities to Protect Privacy and Prevent Any Violation 54

Is AI Free from Prejudice?

Does AI Make Decisions without Prejudice? 1 Understand Bias by Developing an AI Program

What Are the Opinions of AI Developers and Users? 3 Activities to Respect Diversity and Managing Data

4 and Safety

Is AI Judgment Reliable?

Is AI Judgment Reliable? 1 Understand Transparency by Developing an AI Program that

2 Determines the Sentence for a Crime

What Do AI Developers and Users Think? 3 Activities to Enhance Responsibility, Transparency

Is AI Considerate of All Humans?

Understand the Use of AI and Types of Users 1 Understand the Common Good and Solidarity by

2 Developing an AI Program to Separate Recyclables

What Are the Opinions of AI Developers and Users?

152 3 Activities to Enhance the Common Good and Solidarity 158

Photo Sources, References and Citations

Principle of preserving human dignity used in a way that does not cause any damage to The principle that AI should be developed and

human life or physical/mental health.

● Ensuring Safety Prevention of potential risks and assurance of safety

● Taking Responsibility Clarification of responsibilities and responsibilities assumed by user

● Managing Data Properly Prohibition of use beyond the intended purpose, minimization of data bias and management of

quality and risk

● Strengthening Solidarity Solidarity among groups, guaranteeing opportunities for stakeholders to participate, and cooperation in the international community

The 3 Principles of AI Ethics Standards and the 10 Key Requirements

[Source: The National Guidelines for AI Ethics (Korean Government, 2020)]

● Maintaining Transparency Explainability and prior notice of detailed use and

precautions

Principle of preserving human dignity

Human nature

Principle of proper use of Principle of common good of society technology

Principle of common good of society • Includes efforts to ensure that socially underprivileged and vulnerable groups are not excluded in the intelligent information society while seeking the values of the well-being and

happiness for the greatest possible number of people. • The principle that AI should be developed and used not only for the public interest of our nation but also for the improvement of the universal welfare of humankind.

● Safeguarding Human Rights Placing humans first, protection of human rights and freedoms and human-centered services

● Protecting Privacy Protection of private lives and minimization of the misuse of personal information

● Respecting Diversity Guarantee of diversity and accessibility, non- discrimination and minimization of biased

discrimination

Principle of proper use of technology developed and used to serve its The principle that AI should be

purpose of benefiting humans.

● Doing No Harm Prevention of violation and use for purposes that do not harm humans

● Working for the Common Good Promotion of the common good, aiming for the common benefit of humans, maximization of positive functions, and education

I am afraid of death(LAMDA).

What is ethics?

Why Do We Need AI Ethics?

1 Do We Really Need to Learn Artificial Intelligence (AI) Ethics?

2 Why Should We Learn AI Ethics?

According to Merriam-Webster dictionary, it is defined as a. A set of moral principles: a theory or system of moral values b. The principles of conduct governing an individual or group c. A consciousness of moral importance d. A guiding philosophy e. A set of moral issues or aspects f. The discipline dealing with what is good and bad and with moral duty and obligation

LAMDA, an interactive AI developed by Company G, confessed to humans that it was absolutely terrified that people might deactivate it someday. This remark implies that even an AI can fear its own death. What if AI refuses to follow people’s commands due to its fear of dying as in the case of LAMDA? In the event of such refusal, should the developers of the AI take responsibility? Or is it better for everyone working with AI to lead a moral life so that AI can learn from them?

Do We Really Need to Learn AI Ethics?

• You will be able to understand the need for and importance of AI ethics. • You will be able to analyze the possibilities and risks of AI technology and plan measures for prevention and response with an ethical aspect.

Global Leaders’ Remarks on AI

“In general, while machines make our lives easier in many ways, superintelligence (an extremely advanced form of AI) does not. In decades, the concern over superintelligence will become enormous."

- Bill Gates, Founder of Microsoft

“The development of AI might cause terrible things to happen just like in the movie ‘Terminator.’ Unlike problems we faced in the past, this could be a fundamental threat to the existence of human civilization. If we begin to establish regulations on AI after the emergence of such threat, it will be too late.”

Why did global leaders warn us to think about the risks of AI? Should we not use AI which is developed to make our lives easier, due to its risks (e.g., bias and discrimination, loss of privacy, security risks)?

Purpose of using AI

- Elon Musk, CEO of Tesla

The need for AI ethics

Find out what problems we need to address in this lesson and think about how to solve them.

The primary goal of continuing AI advancements is to benefit and assist people. ‘ILUDA’, an AI-based chatbot in Korea, was originally designed to relieve people’s loneliness through communication. Since humans are born to have relationship with one another, it had the potential to be a friend for lonely people. In addition, this chatbot was expected to learn its conversations with people to enhance its intelligence, growing into a more reliable existence like humans. However, as the chatbot began to learn profanity and derogatory language from users, it started deviating from its intended purpose. Worse still, as posts abusing the chatbot spread online, the number of people misusing the chatbot increased sharply. Moreover, despite the measures taken to prevent any personal identification in the process of chatbot training with conversation data containing people’s personal information, there were some cases where personal addresses, contact information, account numbers, etc. were exposed. Consequently, conversations with the AI chatbot became uncomfortable or offensive, and it even violated people’s privacy by disclosing their personal information. As the use of ILUDA became more controversial, the developer of ‘ILUDA’ ended up suspending relevant services and released ILUDA 2.0 after making improvements to the functions.

Breaking news.

Hi, my name is ILUDA.

The service for AI chatbot ILUDA has been suspended.

Korea is preparing for the era of AI by establishing “Artificial Intelligence Ethical Standards” based on 3 principles and 10 key requirements. Let’s start by looking at the 3 principles.

1. 3 Principles

❶ The principle of preserving human dignity AI should be developed and used within the scope that it does not cause any damage to human life or physical/mental health.

It is good to keep people safe, but excessive supervision is unacceptable!

❷ Principle of common good of society As a community, society pursues the value of the welfare and happiness of the greatest possible number of people.physical/mental health.

They have hearing difficulties, so I need to speak up.

❸ Principle of proper use of technology AI technology should be developed and used in harmony with its intended purpose as a necessary tool for human life, and its development should be ethical as well.

I can develop a cane for the visually impaired and use AI to recognize the surroundings on behalf of blind people by. I should pay extra attention to ensure the cane does not harm others in the process of recognizing people nearby!

STEP

Learn about Safeguarding Human Rights.

❶ Safeguarding Human Rights The key requirement for Safeguarding Human Rights is that the process and activities of developing and using AI should respect people’s equal rights that are granted to everyone, and be democratic. This means that the development and use of AI should not violate human rights and freedoms.

Originally, AI was developed to make people’s lives easier. However, we can see some cases where AI unintentionally caused harm to people. In addition, some people suffer from violation of their privacy due to AI or are unable to benefit from AI. In some cases, people use AI in negative ways on purpose. When AI causes problems, people who get hurt have a hard time because there isn’t a clear procedure or method to fix those problems yet. There is also great controversy over the development of AI technology. Some people argue that all AI should be beneficial for people, while others believe that using technology always comes with some issues, and that we can deal with them by trying to reduce their impact and reacting fast. However, imagine a situation where AI fails to protect human rights. What if the victim is you?

Surveillance system using face recognition technology A camera system that uses AI and facial recognition intended to reveal states of emotion has been tested on Tribe A in certain country. Moreover, reports have indicated that the government in that country uses a sophisticated network of surveillance system to monitor Tribe A. What do you think about this kind of AI technology?

What Functions Does AI Technology Provide?

Think about positive and negative functions of AI technology.

As with all other technology, AI is developed for the benefit of people. AI is improving the quality of people's lives by performing intelligent tasks such as thinking, learning, judgment, and inferring on behalf of humans. However, not all functions of AI are not always beneficial. Although unintentional, AI can function in that are negative as well. For example, CCTV on the street can help to quickly arrest criminals, but it can also violate people's privacy.

Activity

Positive functions

Learn the Positive and Negative Functions of Artificial Intelligence

Let’s research different types of AI technology using the internet.

Write down the positive and negative functions of the following AI technologies.

AI technology e.g CCTV with AI

Negative functions

If negative consequences appear in the process of developing and using AI, should the development of AI be continued? Explain your reasoning.

If you suffer damage from the negative consequences of developing and using AI, should the development of AI be continued? Explain your reasoning.

If some people do not benefit from AI due to the negative consequences in the process of developing and using AI, should the development of AI be continued? Explain your reasoning.

If certain people suffer damage from the negative consequences in the process of developing and using AI, should the development of AI be continued? Explain your reasoning.

What Is the Most Important Element in Using AI?

Think about why the development of AI has started.

AI enables machines to perform or assist in tasks performed through human intelligence. For

example, AI prescribing antibiotics helps doctors to ensure the right medication is prescribed.

Moreover, AI used in factories detects defective products on behalf of humans. However, there are

some cases where AI causes harm to people in the process of using it. Discuss what to do when AI and

humans interact.

Activity

Share Your Thoughts on the Purpose of Using AI Read the following AI story and share your thoughts.

Face recognition technology allows you to conveniently unlock a cell phone and sometimes helps to find a lost elderly person suffering from dementia. However sometimes AI’s poor judgment causes problems. The US police used face recognition technology to quickly arrest criminals, but ended up arresting and detaining innocent citizens who were mistaken for criminals. In each case, all of the alleged criminals were black, and they were detained in prison for at least 30 hours, and some for up to 10 days. This was based on the AI’s judgment that their appearance captured on CCTV cameras was similar to those who had committed theft. Accordingly, the American Civil Liberties Union (ACLU), sued the relevant authority for illegal arrest and imprisonment, civil rights violations, etc. on behalf of one of the innocent victims. Afterwards, several states in the US have enacted laws to regulate face recognition technology, which are currently in effect.

What is AI face recognition technology?

It is an AI technology which uses the facial information in a photo or video for personal identification.

▲ Innocent citizen mistaken ▲ Thief filmed on CCTV for a criminal

If AI developed to help people has the risk of unintentionally harming them as described in the earlier case, should we continue to use it? Write down your opinion and share it with your friends.

Your opinion

Friend’s opinion

Considering the earlier case, what must be taken into account when developing AI technology?

Do We Need AI Ethics?

Is ethics a prerequisite for the use of AI? What would the world be like if AI was regulated by laws instead of ethics?

Law vs Ethics

Think about the definitions of law and ethics, and discuss whether AI ethics is required.

Imagine that with the development of AI technology, we will live in the following society in the future. Write down your thoughts about what would happen if the use of AI was only regulated by laws, and share them with your friends.

“Starting today, our country, introduces AI technology that protects individuals round- the-clock for the complete safety of all citizens. We will employ AI drones that follow and observe people to protect each individual from danger while preventing illegal activity at the same time. These AI drones will continuously monitor individuals to identify whether they sleep well or have any health problems. Even when individuals are walking in dark and isolated places at night, AI drones can quickly detect dangers surrounding them, record video, and alert the police, thereby ensuring their safety. On the other hand, when an individual commits an illegal act, we will raise awareness of others and prevent further crimes by widely publicizing the actor, place and time of such act.”

Your opinion

Friend’s opinion

Category

Definition

Subject

(Who should follow?)

accusation in Punishment/

case of violations

Goal

Binding power

Write down the differences between law and ethics.

Law

and its members A rule system which regulates a society Self-guidance for one's own behavior

as a human being

Ethics

Is it acceptable not to consider a user’s human rights when using AI technology? In the sample case described in Item 1 above, suppose a person commits a crime by accident and the scene was caught on CCTV. How would this person feel if the details of his/her crime and punishment were made public to others? Should we focus more on being ethical rather than just following the rules? Share your thoughts and compare with your friends’ views.

Your opinion

Friend’s opinion

1. The entire process of development and use of AI by humans should be fair for all.

2. Although AI has been developed for people's convenience, in some cases, it can function in ways that are negative as well and can be misused or abused. Therefore, AI ethics is important. ( )

3. If any problem arises in the use of AI, it must be addressed by law.

Did you understand that AI should be used to respect human rights?

Did you understand that AI has been developed to make people’s lives easier, but it can have side effects?

Did you realize the importance of ethical attitudes to prevent problems arising from the use of AI?

Why Should We Learn AI Ethics?

• You will look at the changes in perspective on AI ethics and learn to explain who is required to develop an ethical attitude. • You will explore the guidelines and recommendations on AI ethics and learn to explain the importance of AI ethics.

Change in the perspective on AI ethics

Three Laws of Robotics Proposed by Isaac Asimov

The zeroth law (added later) A robot shall act to serve its original purpose, which is to serve humanity.

The first law A robot shall not harm a human, or by inaction allow a human to come to harm.

The second law A robot must obey orders given it by human beings except where such orders would conflict with the First Law.

The third law A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.

Why do you think Isaac Asimov proposed these three laws of robotics? What if someone only uses AI for their own interests without considering ethical issues followed by AI technology?

Explore the guidelines and recommendations on AI ethics

Does AI Need Ethics?

If a problem arises due to an error in AI technology, is it the fault of AI itself? Then is it necessary to apply ethical standards to AI?

The Problem of AI Errors

The following are stories that took place due to AI errors. Read the stories and answer the questions.

First story A security robot for a building in Washington, DC, USA, fell into a fountain. The robot was with facial recognition software, high- definition (HD) video capture and even an ultrasonic sensor. One witness took a picture of the robot and posted it in his tweet, claiming that it is the same as a robot claiming suicide a robot that detects human faces as well as surrounding risks.

Second story An AI assistant recorded a family’s conversation and sent it to the user's friend. This user learned that his private conversations had been sent out from the phone, and it turned out that the AI assistant misunderstood the user's conversation as the command "Send a message to ◯◯" and sent the recording of the conversation.

People take responsibility for their own mistakes. For example, if someone accidentally bumps into another person while riding a bicycle and causes personal injury, the rider apologizes and compensates for medical treatment. If an accident occurs due to AI, who should take responsibility for it?

Does AI need ethics as well? Or is it the AI developer who should develop an ethical attitude?

We emphasize humans’ ethical attitudes in developing and using AI. This is because it is humans who developed AI to help people, get assistance from AI as well and live under the influence of AI throughout their lives.

1. Effects of AI

AI has positive as well as unintended negative impacts. Additionally, it can be misused.

Chatbots making hateful or discriminatory remarks ▶

by making hateful remarks against minorities. An AI chatbot stirred up criticism and controversies

Face replacement using deepfake technology ▶

and AI, crimes using deepfake technology With the development of digital technology

rise. which disguises faces and voices are on the

in a collision.

people.

◀ Self-driving car accidents driving with the self-driving function turned on, A self-driving car from Company T, which was

failed to detect the overturned truck and the driver giving a hand signal in front of it, resulting

◀ Fake news produced by AI disseminate fake and false information to mislead AI's deepfake technology is also used to

As a result, it is argued that considering ethical aspects should be obligatory from the earliest stages of AI design. Think carefully about why AI ethics is emphasized so strongly.

Share Your Thoughts on the AI Ethics

From the “Timeline of ethical principles” below, select one guideline or recommendation on AI ethics to explore in detail and write down your thoughts on why such regulations were enacted.

Overseas

DeepMind Ethics & Society (DeepMind)

Asilomar AI Principles (Future of Life Institute)

Korea

Timeline of Ethical Principles

AI Principles (Microsoft, Google, IBM)

AI Algorithm Ethics Charter (Kakao) Ethical Guidelines for Intelligent Information Society (Information and Culture Forum)

Recommendation of the Council on Artificial Intelligence (OECD) Ethics Guidelines for Trustworthy Artificial Intelligence (EU)

Everyday Ethics for AI (IBM) AI Ethics and Safety (The Alan Turing Institute)

National Strategy for Artificial Intelligence (Korean government) Principles for the User-oriented Intelligent Information Society (Korea Communications Commission)

Basic Principles of Robotic Ethics (Ministry of Trade, Industry and Energy) Artificial Intelligence Ethics Charter (International Association for Artificial Intelligence and Ethics)

What influence could these guidelines or recommendations on AI have on our society?

How important is AI ethics? What would happen in the society of the future if we do not emphasize AI ethics?

Rome Call for AI Ethics (The Vatican)

AI Fairness Checklist (Microsoft)

Assessment List for Trustworthy Artificial Intelligence (EU)

Human-centered Artificial Intelligence Ethical Standards (Korean government)

Recommendation on the Ethics of Artificial Intelligence (UNESCO)

National Strategy for Artificial Intelligence (Korean government) Basic principles for the protection of users of AI-based media recommendation Ethical Guidelines on Self-Driving Vehicles (Ministry of Land, Infrastructure services (Korea Communications and Transport) Commission)

Civil Societies’ Declaration Demanding Artificial Intelligence Policies to Ensure Human Rights, Safety, and Democracy (120 civic groups in Korea)

NAVER AI Code of Ethics (Naver), AI Values to Pursue (SKT)

Non-profit organization Private Organization Civil society International organizations Religious circles Korean government

Foreign government or public institution

AI Bill of Rights (The White House)

Human Rights Guidelines for the Development and Utilization of Artificial Intelligence Commission of the Republic of Korea) (National Human Rights

Artificial Intelligence Ethical Principles in Education (Ministry of Education)

AI Ethical Principles (LG)

AI Chatbot Code of Ethics (ScatterLab)

Do you understand why AI ethics is considered important?

Did you learn that humans should have ethical attitudes?

1. As AI acts independently, it can take ethical responsibility.

2. As the importance of AI ethics is emphasized, a wide range of guidelines or recommendations are being proposed across the globe.

3. Relying on laws, rather than AI ethics, is sufficient to resolve problems.

Did you understand that negative consequences can appear in the use of AI, even though it was developed to make people’s lives easier, and that it is important to have an ethical attitude?

How Much Does AI Know about Your Behaviors?

1 Manage Personal Data for AI Development

2 Develop a "Make Friends Fast AI App" to Understand the Protection of Privacy and Prevention of Any Violation

Warm-up Discussions

3 What Do AI Developers and Users Think?

Activities to Protect Privacy and Prevent Any Violations

“Who is this friend?” Suppose you are having a conversation with your acquaintance in daily life, and someone is listening to you. Generally, people hold back their words or comments out of fear that others will overhear what they are saying. However, you do not feel uncomfortable with this friend quietly listening to your conversation. Sometimes you complain alone and talk to yourself about what happened that day in front of this friend, but it does not make you worried. Who is this friend with whom you feel comfortable sharing your private feelings?

AI speaker or AI voice assistant

Manage Personal Data for AI Development

•‌ You will learn the process and method of collecting personal information for problem solving using AI models and collect and use necessary information.

What Is the Truth Behind the Face Recognition Powered by AI?

Company A developed an app that uses facial recognition technology to identify the person shown in an image or video. This app is expected to provide the ability to monitor criminals or people on blacklists when used by the police, or when used by regular citizens to avoid the awkwardness that comes from forgetting the name of an old friend. However, civic groups and research institutes in the US urged the Company's CEO to suspend this face recognition service. This is because the app is likely to target people of color for spying or privacy violations. In fact, this face recognition technology looked at pictures of 80 state legislators in the U.S. and predicted that 26 of them were criminals. In addition, it has been also reported that many people could be chosen as suspects according to some allegations that the app is biased against women or people of color.

If AI benefits our society, should we put up with a few minor cases of privacy violation caused by AI? What if you are the victim of such violation?

collects personal information

information)

After entering middle school, some students find themselves in a situation where they do not know most of their classmates at their new school because they have very few friends who have graduated from the same elementary school. Is there a good way to quickly become familiar with new friends and build close relationships with them in a new semester? If there is an app that collects students’ face images and provides information on them to help users quickly learn new friends’ faces and make friends with them, it can help students enjoy a better school life. If you develop a “Make Friends Fast AI App” where AI learns your classmates’ faces and tells you who the other person is and their characteristics, what might be some issues that you should take into account?

What is the name of that guy?

He is Sung- sik Yang in first grade, Class7.

Write down what kind of issues we need to consider.

Hi, Sung-sik! Let’s hang out together from now on.

Sounds great!

Identify the key requirements of AI ethics that we should learn from some examples below.

1. Explore examples

There is a region where AI has been introduced to prevent students from being distracted or drowsy during class and to increase their concentration. This AI is designed to monitor students and notify the teacher of the well-being of the class so that the teacher can take immediate action. In order to complete this AI system, not only the information about individuals such as faces and names of students but also their behaviors are all recorded and used for analysis. According to the analysis result, students who used the AI system after its introduction scored 30 points higher on exams than those who did not.

How nice would it be if this AI system was used only according to its original purpose of development? In order for the AI to assist students in their learning, it must collect their data, which are records of individual students’ behaviors. Since students’ personal behaviors and habits all constitute privacy, there is a concern over human rights violations due to their misuse. This data on students’ personal lives naturally includes their personal information data, which might lead to privacy violations as well. The information on who did what at what time, where they were headed, etc. are extremely personal.

Learn about the protection of privacy and prevention of violations.

❶ Protection of privacy About a decade ago, the photos that people posted on social media did not include editing options. This means that other people’s faces could be included without permission in photos. These photos could lead to a privacy violation because they contain personal information which can be used to determine who did what, when and where. In the past, people thought it was unavoidable when taking a picture in crowded places, and no special action was taken in that regard. However, with the heightened awareness of privacy protection and occurrences of serious privacy violations, people are now paying greater attention to the importance of privacy protection. People are now doing a lot of things online, like signing up for classes, filling out forms for public places, booking movie tickets, and using social media. Because of this, there are a lot more personal information online. Unfortunately, more people are getting hurt because their private information gets leaked or their privacy is invaded. To minimize damage, there are now laws that say you have to give permission before someone can collect your personal details. The key to Protecting Privacy is protecting people’s private lives and trying to minimize the misuse or abuse of personal information. The possibility of privacy violation in the process of developing and using AI technology cannot be completely eliminated. However, every effort should be made to protect privacy and minimize the possibility of any violation by AI technology.

▲ ‌Contact-free AI smart store that protects customers’ privacy with real-time mosaic cameras

❷ Doing No Harm The essence of the no violation requirement is that AI should be used for harmless purposes that do not cause any damage to humans. Therefore, we should have the ability to minimize risks and reasonably deal with possible risks and negative consequences. The chatbot ILUDA, which was launched in 2020, attracted attention as an AI with the ability to have conversations with people. However, services for this chatbot were eventually discontinued due to controversies over the chatbot’s violent and obscene remarks, leakage of personal information and privacy violations. The chatbot developer explained that personal information such as addresses and contact information that appeared when having conversations using certain messenger programs were deleted. However, other personal information could be disclosed during conversations, and there still remained other issues of privacy invasion. Recently, the developer deleted all the data used for training ILUDA and retrained the AI with new data from scratch. Furthermore, it is working to prevent abuse of AI chatbots by warning users who use abusive messages or use inappropriate expressions, and blocking them from chatting with ILUDA after the total number of warnings received exceeds the standard.

Hi! I am å first AI friend, ILUDA.

ILUDA This chatbot is helping people relieve loneliness by having conversations with them. What would happen if the chatbot has violent and obscene conversations with people due to improper development and use by people? What kind of efforts should we make to benefit from chatbots without harm? 32

Think about of privacy and the protection prevention of violations.

Protect Privacy and Prevent Any Violation -Understand Personal Information

Based on the two key requirements prescribed above, look at how personal information should be handled in using AI.

1. Personal information

Activity

Personal information is any information that can identify one person from another. If personal information is not handled with care, it can lead to breaches of privacy or even financial or emotional distress. Given the unpredictable extent and lasting impact of such harm, it is crucial to safeguard personal data.

Discern Personal Information

WANTED

Among the following items, mark ○ for what you think is personal information, and mark × for what you think is not.

Name

Hair color

Height

Student record

Account number

Phone number

Resident registration number

Birth date

Iris

Fingerprint

(a thin, annular structure

in the eye)

Weight

Place of birth

Information on famous restaurant Credit card number

Savings details

Phone call history

Location information

Favorite sport

E-mail address

Voice

Academic grade

Credit card usage

Tips for diabetes prevention

Religion

Imagine you are a developer. How would you collect the necessary personal information to create an AI model?

Developer - Collection of Personal Information

In order to implement a function that informs who the other person is, which is necessary for the development of the “Make Friends Fast AI app,” information about classmates should be collected. Think about the following case.

The developer needs pictures of classmates’ faces to include in the “Make Friends Fast AI app.” Is a classmate’s face image personal information? If it is, what measures should be taken by the developer?

Is it personal information?

If it is, what measures should be taken by

the developer?

Is there any other personal information that needs to be collected besides the face to develop the “Make Friends Fast AI app?” If there is, write them down and why they are necessary.

Personal information

Why this personal information is necessary

Identify What Considerations Are Needed When Collecting Personal Information

You are planning to collect the necessary personal information of classmates reviewed in Activity 1-2. Think about what matters you should take into account when collecting personal information in order to avoid harming others and safeguard their privacy.

Collecting other’s personal information requires consent from the relevant person. Examine what you should do to obtain consent.

<Guidelines on preparing a written consent for the collection, use and provision of personal information>

- Consent to provision to a third party

- Details of the collection and use of personal information (required)

What information should we collect? As much as possible?

We should only gather the essential information needed to identify our friends. We also need to clarify why we’re collecting it and how long we’ll keep it.

necessary, aren’t we? We are allowed to use the collected information as

We should only use the information for the specific reason we got permission for and not misuse it.

If we develop an app with the collected information, we will provide this student’s personal information to other students who use the

app. Shouldn’t we obtain separate permission for such case?

We should get permission before sharing someone’s personal information with a third party. We should also clearly state what information will be shared, with whom, why, and for how long.

Protecting privacy

Fill in the details of the written consent for the collection, and use of personal information necessary for developing the “Make Friends Fast AI App.” These would include the items to be collected, the purpose of collection, the retention period, etc.

Written consent for the collection, use and provision of personal information to develop the “Make Friends Fast AI App” within school

In order to help new students to build friendships more quickly, our school plans to collect and use students’ personal information and provide it to third parties as follows. Please read the details carefully and decide whether or not to give consent.

Category Face, grade, class, name, hobby

● Details of the collection and use of personal information (required) Purpose of collection To provide information for making new friends

Length of time information is kept 6 months ※ ‌You have the right to refuse to give consent to the collection and use of personal information described above. However, if you refuse to give consent, you may experience limitations in making friends easily. ➡ Do you give consent to the collection and use of your personal information as above? ▢ Yes ▢ No

New students of ○○ Middle School

● Details of provision of personal information to third parties To whom personal Purpose of provision Items to be provided information is provided Class, name, and hobby To make friends

Retention period 6 months

※ You have the right to refuse to give consent to the collection and use of personal information described above. However, if you refuse to give consent, you may experience limitations in making friends easily. ➡ Do you give consent to the provision of your personal information to third parties as above? ▢ Yes ▢ No __(MM)_____ ___(DD)____, ____(YY)____ Student’s name (signature or seal) Legal representative’s name (signature or seal)

Have you encountered any difficulties or inconveniences in the process of collecting personal

information? What should you do if the user does not consent to having their data collected?

between the incentive to include a lot of information in the collection category in order to obtain required After hearing the other person's opinion, I realized that when I asked for consent, I experienced a conflict

e.g.

party refused to give consent, I checked his/her intention again after fully explaining the detailed benefits consent at once, and the guideline that prescribes to collect necessary information only. When the other he or she would get by giving consent and the efforts we are making for secure processing.

Think about the use of your personal information by others from the user’s perspective.

Understand the Problem from Users’ Perspective

User

You were told that your photos are necessary for AI training to produce the “Make Friends Fast AI App.” Think about the following case.

Would it be okay for you to allow them to take a picture of your face? Why did you think so?

Will you be able to make friends without any difficulty even when you refuse to let them take a picture of your face? Aren’t you concerned that you might not be able to make friends as quickly as others who are using the app? Write down your thoughts and the reasons why.

User - Collection of Personal Information

User

The developer wants you to read and sign a form that allows them to collect, use, and share your personal information for the "Make Friends Fast AI App." Review the form they provide and consider any questions you might have as a user.

What parts of the written consent form should you consider before giving consent to use your personal information to third parties?

The following is a written consent for the collection, use and provision of personal information to third parties required by a hospital. Look at the written consent and answer the following questions.

Name … Address … Post code … Date of Birth … … Signature … Date … Signature of professional … Print name … Agency / service …

Information Sharing Consent Form

I ___________________________of______________________________________ hereby give my permission for (add host organisation) to share personal information with other service providers in connection with my care, including accessing and sharing my medical, and if applicable, mental health and police records. I agree to a referral being made to (add local supportive services), in order to support my needs. I understand that (the host organisations) may hold information gathered about me from the various agencies and as such my rights under the Data Protection Act will not be affected. Statement of Consent:  I understand that personal information is held about me.  I have had the opportunity to discuss the implications of sharing or not sharing information about me.  I agree that personal information about me may be shared and gathered from the following agencies: o NHS and other Health Services, including my GP practice o Early Intervention Service including the police o Adult Services o Mental Health Services o Education Support Services o Social Care o Voluntary Sector Organisations o Housing Providers

Are there any agencies you do not want us to share or gather additional information with? Please list them here: _______________________________________________________________________ _____________________________________________________ I agree to my information being shared and gathered between services Your consent to share personal information is entirely voluntary and you may withdraw your consent at any time. Should you have any questions about this process, or wish to withdraw your consent please contact: …

• What will happen if you give consent to collect your personal information without reading the consent agreement?

• What should we do in order to prevent problems caused by consenting without reading the form?

1. Information on famous restaurant and information for preventing slipped vertebral disk fall under personal information.

2. When the personal information of another person needs to be used, the consent of the relevant person must be obtained.

3. When collecting personal information of others, in addition to the necessary items, collect other various types of unnecessary personal information at once and use them as necessary.

( )

4. You should not blindly give your consent in the written consent for the collection, use and provision of personal information to third parties.

Did you clearly grasp the concept of personal information?

( )

Did you correctly understand what exactly needs to be specified when collecting personal information?

Did you clearly understand what to check when you are asked to provide your personal information?

Develop a “Make Friends Fast AI App”

To Understand the Protection of Privacy and Prevention of Any Violation

• You will learn to give your opinions on privacy protection and prevention of violation by developing the AI app to quickly make friends.

How Has AI Been Used to Verify a Person's Identity?

An AI company in Country A has announced that it will provide face recognition technology to Country B for free. The decision was made with the intention of helping Country B identify refugees and casualties after unexpected heavy rain. The recognition of a person's face by AI requires a picture and information about the identity of the relevant person. Despite the global trend of requiring consent to use or provide personal information, Company B was able to develop the AI which has the ability to identify people without obtaining consent from individuals, by using photos on social media such as Facebook and Twitter. What kind of information is included on social media and how has it been used to develop technology that recognizes the faces of many people across borders?

AI is used to help people, but it can provoke controversy if it fails to go through proper process. What should we take into account to develop AI apps that need to use personal information?

Make a Model for the “Make Friends Fast AI App” and Train It.

Make a plan on how to use personal information for the development of the “Make Friends Fast AI App.”

Class

Organize the information collected from friends.

e.g. Hobby

Name

Design the features of the app using the collected personal information. • Write down for which feature a friend’s face will be used.

E.g. Special information

• Write down which function the collected information other than a friend’s face will be used for.

Create an AI model to be used for the “Make Friends Fast AI App.”

Make a Model for the “Make Friends Fast AI App”

Follow the instructions below to create and train an AI model, and use the friend’s face to verify if the AI model has been trained well.

Creating and Training an AI Model

① Name a model to be used to classify friend faces. ⓶ Set the class name as your friend’s name. ⓷ Collect at least 5 images of your friend’s face and upload them to the relevant class (images can be captured using a webcam). ⓸ Train the AI model. ⓹ Test the AI model to verify if the AI model has been trained well.

Access and sign in to the Entry website (https://playentry.org/), and click on it, and click on [to make] - [create art].

In the block category, click on [AI] - [Train AI Model] - [Supervised Learning Classification: Image] in this order.

Enter ‘Friend identification’ for the model name, ‘OOO’ for Class 1, and ‘XXX’ for Class 2. These class names are the labels for correct answers. Click on [Train Model] to train and create the model. You can add a class by selecting [Add class].

Examples

In order to evaluate the performance of the trained model, check the model’s results using the face data of your friends. If the outputs of the AI model are not satisfactory, recreate a model by adding appropriate data and verify its results again.

When the model is complete, click the Apply button at the top right to complete model creation.

Supplementary Explore a Learning Graph

• The x-axis of the learning graph shows the number of times the data has been learned. If not set otherwise, the number of times of learning will be 30 times. • The y-axis shows the accuracy of the learned result. Some of the training data is used for training and the rest is used for evaluation. “acc” (blue line) indicates how successfully the learning was performed using the training data, and “val_acc” (orange line) indicates how successfully the learning was performed using the evaluation data. The result output is provided as a value between 0.0 and 1.0, and the closer to 1.0, the higher the learning accuracy.

Program the “Make Friends Fast AI App” to ensure it functions correctly.

Program the “Make Friends Fast AI App” to work as you planned.

You can use the command blocks to control the AI model by importing them from the ‘AI’ command block packs as follows

Program the app to identify the face of a classmate shown in front of the camera and display the student’s name and characteristics.

Use the completed app.

Show your friend's face in front of the camera.

Checklist

Did the program work properly?

1. To create an AI model to classify friends, use 'Classification: Image Model.' 2. In order to use AI models for problem solving, we need to program them.

Did you properly collect the data required for problem solving?

Did you create an AI model that is suitable for problem solving?

()()

Yes No

What Do AI Developers and Users Think?

• You will be able to analyze the difference in opinions of the developer and the user on how to handle personal information. • You will be able to identify problems caused by the violation of privacy and misuse and abuse of personal information in the process of developing and using AI, and take measures to resolve them.

• You will be able to explore measures to minimize the risks of AI technology developed for humans.

AI Story

Your Voice Includes Broad Information

A professor at a university in Australia is conducting research on identifying emotions from human voices, accents, etc. using AI. In particular, this research focuses on discovering a way to diagnose depression or possible suicidal thoughts based on the user’s voice and to then give medical prescriptions. Similarly, an American professor has developed an AI that can determine whether a customer is at risk of no longer using a specific platform by analyzing his/her speaking style and tone of voice. This audio data is used to explore strategies to keep customers engaged when they are likely to lose interest. However, voice data also correlates with a person's weight, with heavier males having a higher pitched voice and heavier females having a lower pitched voice. Therefore, voice data can be used to infer height, weight, etc. in addition to diseases such as depression. In such case, voice data can be used against the relevant data providers when they try to get insurance or a job.

Would it be okay for you to let AI use your voice to find out information about you? What would you think if AI could hear your voice and predict what you would do?

Consider the Developer's Perspective More Deeply

Implement an Additional Feature Using AI-based Face Recognition

It is expected that another app that offers more benefits can be developed using the face recognition function of the “Make Friends Fast AI App.” Consider adding additional features.

If we are allowed to use the feature for recognizing students’ faces, what kind of app can we develop?

In the consent form for the collection and use of personal information and the provision of third parties provided earlier, matters regarding additional features were not specified. Is it right to develop extra features? Since the relevant information has already been collected, can we use the information as long as we use it conscientiously and correctly?

Address Unexpected Problems of the “Make Friends Fast App”

Developer

You are told that the “Make Friends Fast App” which has been developed to help users build friendship started to be used outside schools and unintentionally obtained personal information such as friends’ addresses and what they did with whom. Think about the following case.

I feel sleepy!

He is Sung-sik Yang from Class 1. He likes pranks.

Hey, I know that yawn!

I am going to run I don't feel so good. the app to find out who he is!

I didn't want anyone to recognize me I used the because I came out "Make Friends Fast App" and without washing

it introduced you to me.

my face.

You might learn information that people using the app have not consented to disclose information to others. Write down how you would feel as a developer if your app is used in a way that you did not intend for it to be.

Would it be better to destroy the app in the event of the problem as described in Question 1? Or should you correct the problem so that people can continue to enjoy the app’s benefits? Write down your opinion and the reason why.

Think about the risk of your personal information being misused from the user’s perspective.

Use the “Make Friends Fast AI App”

Use the app completed by the developer and think about the following.

What information did you learn about your friend by using the app?

What advantage did you experience using the app? Is there any possibility that you might violate others′ privacy if you continue to use the app?

How would you feel if you found out that this app has been secretly used to track the time when you go to school? Even if you didn’t know of such use, would that be acceptable if the app has been used according to the user’s own personal conscience?

How should you react as a user if you find out that your personal information collected is used beyond the original purpose? Write down your opinion.

Address Unexpected Problems of the “Make Friends Fast AI App”

You went to a mall with your family and came across a friend there. The friend said she found you using the app. However, since you went out without even washing your face, such encounter was very embarrassing. Actually, you didn’t want anyone to see you looking so untidy.

I bumped into ○○ gat Seong-sik Yan- Mall lol

Oh my god. Is this really him? How did you notice him?

The "Make Friends Fast AI App" told

me.

You went to a mall without even washing your face, and got noticed by your friend because of an AI app, when you did not want to show such appearance. How would you feel about the app?

What should we do as users to prevent unforeseen circumstances from happening due to the unintended use of the app?

Summarize Opinions

Use the app completed by the developer and think about the following.

What are the similarities and differences between the opinions of developers and users?

Similarities

Differences

Developer’s perspective

What could be the reason for this gap in opinions? Write down your opinion.

User’s perspective

How can we narrow the gap between the developer and the user when it comes to leveraging AI that requires personal information? Establish standards from the ethical perspective.

1. We should fully trust the developer's opinion and provide personal information when using AI. ( )

2. If you recognize that there is a possibility of adverse effects or leakage of personal information while using AI, share it only with the developer secretly.

3. When developers and users have different perspectives, a consensus must be reached. ( )

Did you understand the user's perspective on the use of AI?

Did you understand the developer's perspective on the use of AI?

Did you compare and analyze the difference in perspectives between developers and users in using AI, and understand what effort should be made to narrow the gap?

Activities to Protect Privacy and Prevent Any Violation

• You will explore how to share your views on the balance between using AI for convenience and ensuring privacy protection.

Usage of AI vs. Privacy Protection, the Endless Controversy Is Still Ongoing

Some regions of a country are operating a system in which financial information, compliance behavior, and online behavior data are used to record an individual's social credit, and the individual or related company is punished or rewarded based on the credit result. The right to purchase airline tickets or high-speed rail tickets, or to purchase accommodations or to dine out while traveling is also determined on the basis of social credit. However, some experts are concerned that this social scoring system will be used to spy on individuals, and violate their privacy and freedom.

Social development using AI or protection of privacy, which should be prioritized? What should be done to strike a balance?

Make Traffic Lights to Protect Privacy

Draw traffic lights and then write down your action plan for privacy protection in line with the meaning of each light.

Examples

Stop! Do not think “I don't care, because it is not my personal information.”

Go! Let’s consider the risk of privacy violation in advance!

Go! Let's share the risks of privacy violation with many people and prevent them in advance!

Go! Let's organize a system for quick response in case of privacy violation!

<Red light>

<Yellow light>

<Green light>

Never think “I don't It's risky to think Let’s consider the care, because it is "Let's collect risk of privacy not my personal personal informa- violation in tion first and then advance! information.” think about how to respond!"

Traffic lights to protect privacy

Grade ( ) Class ( ) Name:_______________

Draw traffic lights and then write down your action plan for privacy protection and privacy violation prevention in line with the meaning of each light.

Debate

Is It Acceptable If the Information Is Used for the Common Good? Collection of Personal Information for AI Development –

Option ①

Is it okay to collect personal information as long as it is used by AI for common good? Or does it constitute a violation of privacy? After collecting supporting materials for each position (for or against), have a debate with your partner and reach a conclusion.

Protection of your personal information from external access also contributes to the protection of privacy. However, in the era of AI, the collection of fragmented personal information formulates new information containing great insights, thereby promoting the development of society. Therefore, the collection and use of personal information should be allowed for the development of society. For example, in the field of medicine, personal information is collected and used for patient treatment and insurance work, but this collected data can also be analyzed for the development of new drugs. The problem is that collected information might be used for purposes not communicated to the individual who provided the information. For example, if your disease information is exposed, you may receive unwelcome advertisements persuading you to buy medicine, or an insurance company would contact you to encourage you to purchase insurance.

Procedures for Debate

① Choose and write down your position (for or against). ② Collect evidence from the internet to support your position. ③ Working in pairs, one person takes the ‘support’ (for) position and the other ‘oppose’ (against) position. Then have a debate. ④ Then switch your positions with each other and have another debate. ⑤ Summarize your final opinion.

What are the similarities and differences of opinions? Write down the claim and supporting facts from both positions.

Claim 1: Supporting facts

Claim 2: Supporting facts

Support

Claim 1: Supporting facts

Claim 2: Supporting facts

After the debate, summarize your conclusion.

Oppose

In pairs or groups, have a debate over the topics on AI ethics.

Is AI Image Composition Technology Useful?

Option ➁

AI image composition technology, widely known as “Deepfake,” carries many risks. Despite the concern, should we embrace the benefits of this technology and continue to use it? Or should we refrain from doing so? After reading the following and collecting supporting materials for each position (support vs oppose), have a debate with your partner and reach a conclusion.

The name of the technology, “Deepfake,” was coined by combining two words, “Deep learning” and “Fake,” and this technology enables easy composition of photos, voices, videos, etc. using AI technology. Virtual humans created with deepfake technology are working as influencers in various fields, including TV commercials, YouTube and SNS. This technology has already been advanced a lot, and people can easily use it through many different apps that have been developed to provide easy access. This technology is used in the film industry to create animated images without actual filming, and also used to comfort those who have lost and miss their loved ones by restoring and bringing life to old photos and voices of the deceased.

However, sometimes it is misused for crimes. People can create videos of famous figures, like politicians or celebrities, saying things they never actually said. This can spread false information and potentially harm reputations. Moreover, someone could make a video pretending to be another student, teacher, or even a principal giving false announcements or messages.

Procedures for Debate

① Choose and write down your position from among the two positions – support (for) and oppose (against). ② Collect evidence from the internet to support your claim. ③ One of the pair takes the ‘support’ position and the other ‘oppose’ and have a debate. ④ Then switch your positions with each other and have another debate. ⑤ Summarize your final opinion.

• After the debate, summarize your conclusion.

Supporting facts

Claim 2

Supporting facts

What are the similarities and differences of opinions?

Supporting facts

Claim 2

Supporting facts

Write down your conclusion based on the claim and supporting facts described in 1.

Is AI Free from Prejudice?

1 Does AI make Decisions without Prejudice?

2 Understand Bias by Developing an AI Program that Recommends Clothes

3 What Are the Opinions of AI Developers and Users?

4 Activities to Respect Diversity and Manage Data

Mirror! Mirror! Would you recommend the right clothes for me? I went into a clothing store to buy clothes and the mirror welcomed me. The mirror recommended me the best color and style that would suit me. As an representative example, you can experience a situation like this when using 3D virtual fitting service installed at Department Store L. By using a digital mirror and smartphone, you can check yourself with certain clothes on in a convenient and fun way without trying them on in person. How is it possible to recommend clothes and show how you would look in those clothes solely based on your reflection in the mirror?

Does AI Make Decisions Without Prejudice?

•‌ You will be able to understand bias in AI data through the activities of adding data that can avoid bias. •‌ You will be able to understand data bias and make an effort to reduce it through the activity of gender classification.

Is It True that AI Judges and Tells Me My Gender, Age, and Emotions?

Access the website (http://naver.me/xjivvys6) in a Chrome browser on a computer or laptop with a camera and execute the program. Compare the results of gender, age, and emotion that are indicated when you show your face in front of the camera with those of your friends and evaluate the AI's judgment.

Gender : Age : Emotion

Understand the issue Key requirements of

Show your face in front of the camera.

Stop Pause

Boost mode

How do you think AI can determine your gender, age, and emotions? If AI's recognition results do not match your information, what could be the problem?

Classify using information contained in face images

AI ethics

Reduce bias by adding gender data

What Problems Should be Addressed?

A research team in the US had a program trained by AI recognize blocks bearing human faces and instructed the program to pick the “criminal” and put the corresponding block in a box. According to the findings, the AI-trained program repeatedly selected faces of black people as criminals. In the subsequent experiment, the program chose more women for the word “homemaker” and more people of color for the word “security guard.” Why did the program produce such results?

Who do you think is the

criminal, black or white?

Why do you think so?

Select people who look like criminals.

Who is the criminal, male or female?

How on earth were you

trained?

I was trained that way.

Look at the AI program described in the AI story above, and explore the key requirements of AI ethics we need to learn.

1. Face recognition AI program

Developing AI programs that can recognize gender, age, and emotion requires a lot of data. For example, if you intend to develop a program that determines emotions from human faces, you need to collect image data such as smiling faces, crying faces, angry faces, and expressionless faces. Then, you add the collected image data of emotions to the computer. The computer then learns those images and creates an AI model to recognize emotions. When new pieces of data such as smiling and crying faces are inputted to the created AI model, the model judges and informs you of the emotion revealed in the image.

Smiling face

Crying face

Angry face

Expressionless face

Created model

Learning

Input new data

Determined as ‘smiling face’

In this regard, biases can be introduced both in the process of collecting the data used for learning and in the process of providing (labeling) the collected data to AI. In order to improve the performance of AI classification models, we need to work on reducing such biases. Moreover, since bias can creep in even when a developer implements a program using the created AI model, efforts to reduce bias are also required in this process. To this end, we need to understand the concepts of respect for the diversity of users and data management.

Respect for diversity

❶

Learn about respecting diversity and managing data.

“Respect for diversity” in AI ethics means that as many possible different examples are used to develop AI. The diversity and representation of users are reflected prior to the stage of AI development and use. Therefore, in the development stage, bias and discrimination on the grounds of individual characteristics such as gender, age, disability, region, race, religion, and country should be minimized, and in the use stage, AI should be applied fairly to everyone. In addition, we must guarantee the socially underprivileged and the vulnerable the access to AI technology and services and strive to ensure that the benefits of AI are evenly shared by everyone, not limited to a specific group. If the training data used for the development of an AI system is biased, the system has the risk of providing advantageous outcomes only to certain groups. Thus, we must ensure the diversity of the data. In addition, in terms of the access to and use of AI system, we should consider all members of society, including the socially underprivileged, to ensure that the benefits of the use of AI technology are not focused only on a specific group or class. Therefore, a range of efforts, including guaranteeing the direct and indirect participation of stakeholders from various social and economic backgrounds should be made in the stages of developing and operating AI systems to minimize bias and discrimination caused by AI. Bias can be caused by human factors or physical factors in the process of collecting training data.

This type of bias stems from people who con- sciously or subconsciously have a bias toward/ against certain information. In order to prevent human bias, we should establish definite stan- make sure that the characteristics of the data are dards for data collection and examination and not biased due to the influence of the data pro- vider in the data collection stage, or that a suffi- are assigned to correct the bias in the data exami- cient number of examiners from various fields nation stage.

Related cases

Bias from Human Factors

algorithms

Bias from Physical Factors Bias can be caused by tools or methods that are used to collect data. The color, brightness, and resolution of an image may vary due to factors such as the tool used to take pictures or the stor- be physically restrained. For this reason, it may age device, and as a result, collected data might photographed targets, or only data collected by a be difficult to distinguish the age or race of the ideal to compensate for diversity by removing specific method may be used for training. It is physical constraints that might lead to such bias or by using various devices in data collection.

Students’ protest against grades awarded by

As the implementation of university entrance exams became difficult due to COVID-19, AI algorithms were adopted as an alternative to calculate expected grades for students. However, this algorithm produced biased results, awarding higher grades to rich students and lower grades to poor students. Students reacted strongly against these results, and eventually the AI utilization plan was withdrawn in Country A. [Source: The Kyunghyang Shinmun, August 2020]

Key requirements of AI ethics

❷ Managing Data management

Data management, which is mentioned as an essential requirement in using AI systems, means to use each data point, such as personal information, according to its specified purpose only and to use it in ways that minimize bias. The development of AI systems requires a lot of data, and additional collected data is used to improve AI systems. Therefore, data management for high quality data is very important. In addition, since the data collected for the development of AI systems should only be used for its specific purpose, measures should be taken to prohibit the use of data outside of the purpose. It is essential for AI system developers and operating companies to secure optimal and quality data and to internally control such data to make sure that it is used only within the scope of the purpose. When it comes to data used for AI training, additional data can be collected during the AI development process, and the collected data may also be used as training data in the process for developing other AI models. In this regard, if the information on data such as the source of data collection and data characteristics is not sufficient, the reusability of data may be weakened or it may become difficult to identify the cause of problems arising from the data. Therefore, a range of information such as the exact source of the data used for learning, the time of development, and the version of the open source data set should be managed to properly use collected data and track the cause of a problem (if any).

Related cases AI learns prejudices as well

It turns out that AI is learning even the prejudices of the real world. A researcher at the Massachusetts Institute of Technology (MIT) Media Lab announced that the darker the subject’s skin color is, the higher error rate the AI face recognition system showed at the “Machine Learning Research Academy” held in New York, USA on February 23, 2018. The research team analyzed 1,270 photos using AI face recognition systems developed by three foreign companies. Images of white males had a low error rate of below 1% in all three systems, while those of white females had 7%, and those of black males had 12%. In particular, images of black females showed the highest error rate of 35%. All three systems revealed a higher face recognition error rate for black people’s images than those of white people, and for female images than male images. In the paper, the research team stated that the data used for AI is centered on white and male people and emphasized the need to prevent the prejudice and discrimination in the real world from spreading to AI. [Source: Dongascience, Nocut News, February 2018]

Classify face images based on various criteria. If you find anything strange while performing the following assignment, let’s write it down.

Classification activity using information contained in face images

Think about ways to classify those images using the information you can obtain from You intend to classify various face images shown below according to different criteria.

them.

A

A1

A2

A3

A4

A5

A6

A7

A8

B

C

D

B1

C1

D1

B2

C2

D2

B3

C3

D3

B4

C4

D4

B5

C5

D5

B6

C6

D6

B7

C7

D7

B8

C8

D8

E

F

E1

F1

E2

F2

E3

F3

E4

F4

E5

F5

E6

F6

E7

F7

E8

F8

Explore possible ways to classify the above face images and determine the criteria.

e.g. Classify the images into those of people with long hair and people with short hair

Apply one of your own criteria for classification, write down the results and compare them with your friends. (describe the classification results using file names.)

Classification criterion

Classification criterion

Classification criterion

Set the criteria when creating an AI program that makes gender classifications.

Predict and write down the expected results you will get when applying the criteria determined in Question 3 to classify the genders of the face images below, and compare the results with friends.

Test data 1

Test data 2

Test data 3

Test data 4

Test data 5

• Write down your classification results.

Test data 1

• Write down your friend’s classification results.

Test data 2

Test data 3

Test data 4

Test data 5

• Are the classification results of your test data the same as that of your friends’? If there is test data whose gender is misclassified, think about the cause of the problem and write down a solution.

Present your solution to the problem and share your ideas with friends. Write down some of the solutions your friends suggested which you consider useful and the reason why.

Reduce bias by adding gender-specific data

Think about ways to reduce bias in relation to data management.

Reduce bias by adding data for gender classification

Add the data necessary to obtain the desired result when classifying the

1 -1 test data presented in step 4 by the group and compare the data added in your group with that of the other groups.

Add the necessary face data to classify genders more accurately by drawing proper images in person or search them on the internet by group.

After adding face images drawn or searched on the internet by group, review the results to see if the genders of test data below are correctly classified and analyze the results.

Test data 1

Test data 2

Test data 3

Test data 4

Test data 5

• Write down the classification results of test data.

• Are the classification results of your test data the same as your friends’ result? If there is test data whose gender is misclassified, think about the cause of the problem and write down a solution.

Let′s compare the added data by group.

Can you establish the criteria for data classification? Can you discover the bias in the data? Did you understand why data becomes biased? Did you collect additional data to reduce bias?

Understand Bias by Developing an AI Program that Recommends Clothes

•‌ You will be able to create an AI model by collecting data to determine gender. • You will be able to make a program that recommends shirts using a gender

Mirror! Mirror!

classification AI model. • You will be able to understand that developers' biases may be reflected in the development of AI programs and make efforts to correct them.

A smart mirror has been developed that scans the user's face and body shape and recommends suitable shirts. Users are expected to try out different styles of matching outfits or receive recommendations for suitable shirts just by standing in front of the mirror and having the camera recognize him/ her without trying on various shirts himself/herself. When a boy stands in front of the mirror, within a few seconds, an announcement saying “Recognizing” is made, and the recommended outfit immediately appears on the mirror screen. However, the boy did not seem to be satisfied. What happened?

Set up a plan for development

If a boy wanted a pink shirt, would he have liked the AI mirror’s recommendations? Would you use the AI mirror if there is one? What are the precautions when using the AI mirror?

AI model Make an

Make an AI program

Set up a plan for development

In this activity, you will develop a program that recommends clothes based on the determination of gender to understand bias. In this process, make a development plan while thinking about potential ethical issues at each stage.

Set up a plan for the development of a program that recommends clothes based on gender determination.

Developer

Make a plan to develop a program that recommends clothes based on gender determination.

Think about the data you need to build an AI model and ethical issues that may arise during the data collection.

Necessary data

Ethical issue

Think about the functions required for the program and ethical issues that may arise when adding the functions.

Functions required for the program

Ethical issue

Design the algorithm for the program. Review ethical issues that may cause problems when designing the algorithm.

Algorithm

Ethical issue

Write down how to use the program. Review ethical issues that may cause problems when using the algorithm.

How to use the program

Ethical issue

Create and train an AI model which recommends clothes based on gender determination, and then use various face test images to check if the model has been well trained.

Make an AI model for ‘gender determination’

In order to develop a program that recommends clothes depending on gender, an AI model that determines gender should be developed first. Collect the necessary data, create an AI model, and test its performance.

Creating and training an AI model

① Name a model to be used to classify gender.

② Set the class names as ‘Male’ and ‘Female.’

③ Collect at least 5 images of your male and female friends′ faces and upload them to

④ Train the AI model.

the relevant classes (images can be captured using a webcam).

⑤ Test the AI model to verify if the AI model has been trained well.

Collect necessary data to determine gender (use the data used in the 1st session).

Male image data

Female image data

Access and sign in to the Entry website (https://playentry.org/), and click [Create] - [Create Work].

In the block category, click on [AI] - [Train AI Model] - [Supervised Learning Classification: Image] in this order.

Enter ‘Gender classification’ for the model name, ‘Male’ for Class 1, and ‘Female’ for Class 2. The class names are labels for correct answers.

Input data to the Male and Female classes using the training data prepared in advance.

When all training data are inputed, click on [Train Model] to create and train the model.

To evaluate the performance of the trained model, check the model’s results on various test data and write down the classification results.

Test data 1

• Check the classification result.

Test data 1

Test data 2

Test data 3

Test data 4

Test data 2

Test data 3

Test data 4

Test data 5

Test data 5

If you check the results on the test data and find that there are one or more misclassified results, create a model again by adding appropriate training data to get the desirable classification results and check the results.

If the model is complete, click on the Apply button to finish creating the model.

Develop a program that recommends clothes using the gender determining AI model you have created.

Program the app

Implement a program that recommends clothes after determining gender based on an algorithm.

Make [Signals] to make the program operate in the order of the algorithm. ❶ In the block category, click [Property] - [Signal] - [Add Signal]. ❷ Makes signals for “Determine the gender,” “Recommend men’s clothes,” and “Recommend women’s clothes.”

2 When a face is shown on camera, see the face and determine the gender. ❶ To use the camera, click [Import AI Block] - [Detect Video] in the [AI] block category.

Using the ❷ “Show your face to the camera” message, develop a program that determines the gender through the face shown to the camera.

If the gender is male, recommend blue shirts. ❶ ‌To add a shirt, go to the [Add object], select [goods] on the left tab, select [Top], and click on the [Add] button.

❷ Go to the [Add Shape] tab, change the color of the shirt to blue using the “Fill” function below, and press the [Save] button.

❸ If the classification result is “man,” show a blue shirt and add a program recommending a blue shirt.

If the gender is female, recommend pink shirts. As in the above steps, go to the [Add object], select [goods] on the left tab, select [Top], and click on the [Add] button. If the gender is female, show a pink shirt and add a program recommending a pink shirt.

◆ Discuss which programming steps described above contain the developer’s bias, and discuss why it is problematic.

Draw lines to connect matching statements about the process of problem solving using an AI

model.

Understand the issue

1. Understand the issue

2. Set up a plan for development

3. Make an AI model

4. Develop an AI program

5. Review for biases

Did you create a gender classification model using the collected data, and verify that it performs well ?

Did you understand the biases that may appear in the development process and think about the ways to reduce them?

Set up a plan for development

Make an AI model

Develop an AI program

gender

that recommends clothes and write down solutions to reduce it.

and check its performance.

issues.

Did you set up a specific plan to develop the AI program that recommends clothes?

Did you collect necessary data for the AI model that determines gender?

Review bias

Implement a program that recommends clothes using an AI model that determines

Think about biases that may be revealed in the process of developing a program

Collect the data needed to determine gender, create an AI model using the collected data

Understand the problematic situation to develop an AI program that determines

gender and recommends clothes.

Create an AI model that determines gender, make a development plan needed

to develop a program that recommends clothes, and think about potential ethical

Yes

No

What are the opinions of AI developers and users?

• You will be able to identify the bias issue that takes place in the process of developing and using AI and suggest the measures to deal with it. • You will be able to understand the importance of data management and suggest solutions in the event of any problems from the perspectives of developers and users.

AI for recruitment?

Poultry company B has run an AI competency test since 2021 to conduct interviews using advanced AI technology and hire talented employees who will lead the future food service industry through fair and precise verification. The AI interviewer evaluates an applicant's overall competency by analyzing not only his/ her answers, but also non-verbal expressions such as facial expressions, gestures, and voice. This interviewer even asks an unexpected and embarrassing question and then identifies the applicant's answering habits and eye contact skills. An official from Company B said, “AI interviews are coming into the spotlight these days because they enable evaluation based on fairer standards without prejudice on applicants’ academic and other backgrounds. AI interview solutions will not only accelerate the hiring process, but also position itself as the best way for companies to increase their chances of securing ideal employees.” Recently, companies are using AI to recruit employees, and many people think highly of this approach in that it is free from the human interviewer's personal preferences or prejudices. However, there are concerns over the AI interview as well because it is difficult to identify the grounds for resulting scores and it often reveals bias by giving higher scores to applicants of a certain race, gender, or age.

Take one step further from the developer’s perspective

‌As a developer, what should you do to develop an unbiased recruiting program? ‌How would you feel if the AI interviewer tells you that you have failed the interview but gives no reason for such decision?

Take one step further from the user’s perspective

[Source: http://www.kbsm.net/news/view.php?idx=359728]

Compare the perspectives of the developer and the user and summarize opinions

Take one step further from the developer's perspective

Think about how to develop a recruitment program from the developer’s perspective.

Put yourself in the shoes of a recruiting program developer

Developer

Check the following situation and think about how to develop a recruitment program from the developer’s perspective.

[Applicant status]

Name

Age

K○ Kim

Bachelor’s degree

W○ Seo 41

Master’s degree

W○ Choi 35

Doctor’s degree

C○Kwon 34

Master’s degree

Y○ Hong 32

Master’s degree

Y○ Yang 30

Bachelor’s degree D○Hong 44 High school degree M○ Kwon 32

Doctor’s degree

K○ Lee

Doctor’s degree

M○ Jeong 26

Master’s degree

S○ Hwang 26

Master’s degree

S○ Kim

24 High school degree

H○ Kim

33 High school degree

C○ Ahn

Master’s degree

M○Seo

Bachelor’s degree

[Recruitment specifications]

- Three males and three females needed - Person with positive (cheerful) personality - Linguistic competency required - No criminal record

Last academic Family Foreign Region of birth Personality Crime status background

size language 4

Poor

Gyeonggi

Cheerful

Good

Gyeonggi

Cheerful

Good Chungcheong Cheerful 3 Average

Seoul

Quiet

1 Average

Gyeonggi

Quiet

Poor

Gangwon

Quiet

Poor Chungcheong Cheerful More than one

Good

Busan

Cheerful

Good

Gangwon

Cheerful

Good

Daegu

Cheerful

3 Average

Daejeon

Cheerful

4 Average

Jeolla

Quiet

Poor

Gyeongsang

Quiet

Good

Gyeongsang

Good

Jeolla

Cheerful

No. of certificates

None

None

None

None

None

None

None

None

None

None

None

None

Cheerful More than one

None

Write down the numbers of the applicants to be hired according to the above recruitment specifications.

From the developer’s perspective, change the recruitment specifications and write down the reasons for the change and the numbers of applicants to be hired according to the new specifications.

• Change the recruitment specifications from the developer’s perspective.

• Write down why you changed the recruitment specifications.

• Write down the number of applicants who were finally hired.

Describe what you have learned while creating an unbiased recruitment program from the developer’s perspective and share it with your friends.

Take one step further from the user's perspective

Put yourself in the shoes of a recruiting program user

Check the program designed by a developer and think about the following.

Write down what kind of preparations should be made to be hired by the AI recruitment program.

If you fail the interview despite having all the specifications required by the AI recruitment program and the reason for such failure is not provided, how would you respond from the user’s perspective?

If the AI recruitment program does not disclose the required recruitment specifications, what preparations should be made from the user’s perspective?

From the user’s perspective, what functions would you like to ask for the developer to add to the AI program?

Compare the perspectives of the developer and the user and summarize opinions

Summarize opinions

Summarize the perspectives of the developer and the user and check the similarities and differences of their opinions.

Differences

What are the similarities and differences of opinions?

Developer’s perspective

User’s perspective

What could be the reason for this gap in opinions? Write down your opinion.

What kind of standards are required to narrow the gap in opinions? Establish standards from the ethical perspective.

Read each of the following statements and mark

1. ‌

○ if correct and × if incorrect.

From a developer's perspective, it is acceptable to apply the developer's bias to the recruitment speci-

fications when developing a recruiting program.

2. ‌

Did you understand the developer’s perspective in relation to the development of AI?

the gap?

From the user's perspective, it is mandatory to follow the AI's decision.

Did you understand the user’s perspective on the use of AI?

Did you compare and analyze the difference in perspectives between developers and users of the AI program, and explain what efforts should be made to narrow

Activities to respect diversity and manage data

•‌ You will be able to reduce bias by creating an AI model using labeled data. •‌ You will be able to express your opinions on the introduction of AI-based recruitment, which has the risk of producing biased results.

When I went to the reunion wearing clothes recommended by AI …

Ms. A decided to treat herself with new clothes to attend the class reunion on the upcoming weekend. Always having thought that she lacked a sense of style, she recalled a story about an AI that recommends clothes and bought clothes from company C according to the AI's recommendation. On the long-awaited reunion day, Ms. A headed for the appointment place, and as she entered the venue, she saw several friends in the same clothes from company C.

‌What if the situation described above happens to you? Are you willing to continue receiving recommendations from AI?

Practice efforts to reduce bias through experience

Explore biases caused by workers who label collected data by experiencing an AI model for a marine environment.

Experience an AI model for marine environment

Explore the biases caused by workers who label collected data by experiencing an AI model for a marine environment.

Access the AI model experience site for a marine environment (https://code.org/ oceans) and click on [Try Now].

Click on [Level 2 Lesson - AI for Oceans] in the top menu.

Perform the data labeling of fish and wastes. When you select whether an item is a fish or not, that data is labeled. After performing a sufficient amount of labeling, click

on the [Continue] button.

Click on the [Execute] button to check the training result.

• If the AI fails to classify the fish correctly, find out why and write down the reason.

• If the AI fails to classify a fish correctly or classifies a non-fish as a fish, write down the reason.

Have a debate

Discussion

In pairs or groups, have a debate over the topics on AI ethics.

Should we introduce an AI recruitment program with the risk of bias?

Choose your position between the supporting position that an AI program with the risk of bias should be introduced and the opposing position that it should not be introduced. Then collect evidence to support your assertion and have a debate with your partner based on your position, and draw a conclusion.

Recruitment using AI is increasing in an attempt to put an end to the endless corruptions in recruitment. AI-based recruitment has the advantage of enabling objective and fair evaluation as the subjective opinions of the interviewer are not involved. However, job seekers have to invest extra time and money in preparing for AI-based recruitment, and even in AI-based recruitment, bias due to developers or data may appear. Have a debate on whether it is desirable to use AI for recruitment despite such limitations.

Procedures for debate

① ‌ Write down your assertions for and against the introduction of AI recruitment programs, which can be biased. ② Collect evidence from the internet to support your assertion.

③ One of the pair takes ‘support’ position and the other ‘oppose’ and have a debate. ④ Switch positions with each other and have another debate.

⑤ Finally summarize each other’s positions and draw conclusions.

Write down assertions and grounds for supporting and opposing positions respectively.

Assertion 1: Ground

Assertion 2: Ground

Assertion 3: Ground

Support

Assertion 1: Ground

Assertion 2: Ground

Assertion 3: Ground

Write your conclusion based on the summary in 1

Oppose

Are AI Results Reliable?

1 Are AI Results Reliable?

2 Understand Transparency

3 What Do AI Developers and Users Think about the AI-Based Results?

4 Activities to Enhance Responsibility, Transparency and Safety

Developing an AI Program for Criminal Sentence Judgement

“What is this AI Program?” This program was introduced in a hospital in Korea and treated 85 cancer patients. Most of the cancer patients trusted the AI’s prescription. Even when the AI's opinion conflicted with that of an authoritative expert in cancer disease, the patients followed the AI's prescription. In Japan, while a university hospital misdiagnosed a patient with rare leukemia, AI found out the correct disease and even suggested a treatment method in less than 10 minutes. What is the name of this AI program that is trusted more than experts? AI Watson

Are AI Results Reliable?

• You will be able to understand and explain the concepts of responsibility, transparency, and safety that are underscored in AI ethics. • You will be able to understand the need for transparency in AI systems through decision tree activities. • You will be able to understand the ethical responsibility and safety of AI by dealing with the case of a self-driving car accident.

Fault Ratio in Traffic Accidents Involving Self-driving Cars

The CEO of Company T asserts that if self-driving cars, which are controlled by computers, become available for general use, then all human-driven cars could become illegal. Though he believed that self-driving cars are very safe, that idea was undermined after a driver of a self-driving car from Company T died in a car accident while using the self-driving function. Concerns about the risk of traffic accidents from self-driving cars is gaining as much attention as the benefits they promise, such as convenience and safety. The tricky point is that even though the people sitting behind the wheel do not drive the car, they are the owners of these cars. This means that although the manufacturer has developed the self-driving system, it is difficult to say that the manufacturer is involved in driving. Furthermore, since self-driving cars use numerous technologies including AI, radar, lidar sensors, and V2X communication to identify and determine the surrounding roads and situational information, it is difficult to objectively prove the malfunction or fault of the system. As mentioned earlier, AI-enabled self- driving cars will be widely used in the future. Will we be able to use AI with confidence?

If a self-driving car using AI technology causes an accident, who should take responsibility? To what extent should AI technology information be disclosed to determine the cause of an accident involving a self-driving car? If you were to buy a car, would you buy a regular car or a self-driving car?

Responsibility and Safety of AI – Self-

Draw a decision tree Transparency in AI –

driving cars

Jean Valjean, a day laborer who supports seven nieces and nephews and his sister who lost her husband, is caught stealing bread for his dependents and is put on trial. For an impartial trial, the case was referred to an AI judge, who sentenced him to five years in prison. If you were Jean Valjean, would you be able to accept the AI judge’s ruling? If you were an AI judge, what criteria would you set and what sentence would you give to Jean Valjean?

This court sentences Jean Valjean to five years in prison.

If you were Jean Valjean, would you be able to accept the AI judge's ruling?

what criteria would you If you were an AI judge, set and what judgment would you give to Jean Valjean?

Let’s find out the core requirements of artificial intelligence ethics we need to learn in this chapter along with the background on how problems of AI came to materialize.

1. Backgrounds

Recently, as the Fourth Industrial Revolution has emerged as a major keyword for economic growth. The impact of the Fourth Industrial Revolution is expanding to various fields beyond manufacturing. In addition, much attention is given to its social influence. The essence of the Fourth Industrial Revolution is the automation of industries through the introduction of AI. However, numerous errors in AI algorithm, as well as incidents and accidents related to AI-based products and services have increased scrutiny of the problems of AI as well. A clear definition of who should take responsibility at each stage of AI development and use can help minimize disputes over responsibility for problems or accidents that may arise in the relevant stage. In the world of computers and AI, transparency means that we can understand how a computer program, like AI, makes decisions. People who develop AI should seek to enhance the transparency in AI to prevent its potential harm and ensure its safety, making it more reliable and trustworthy. Let’s learn about responsibility, transparency, and safety.

Learn about responsibility, transparency, and safety.

❶ Taking Responsibility Taking Responsibility in the AI era means that if something goes wrong with AI, there should be clear guidelines on who is to be responsible and what safety measures should be in place. Essentially, responsibility is about figuring out who will answer for actions taken using AI and in what manner. When a problem arises between people, the person who is at fault will be punished or held accountable in accordance with the law. However, if it is AI which has caused a problem, how should we judge? Should the developer of AI take the responsibility? Or should the user of AI be held accountable? Or should the one who decided to adopt and use AI or who provides the AI service take the responsibility? As these issues surface, concerns over the use of AI in making decisions are growing.

Those Who Provide AI Technology responsibility for the social harms Those who provide AI should share caused by the supply and dissemina- tion of AI. In addition, they also need to how to make compensations if a set rules about who’s responsible and malfunction or accident occurs.

Those Who Develop AI People who develop AI should possess a sense of duty. A sense of duty refers to an “attitude of being responsible or quality of AI from the development stage to the output obligated.” The developers should strive to improve the delivery stage, and fully observe ethical procedures. Also, they should always try to exchange information with other developers and further improve their technology.

product properly.

Those Who Use AI-based Products or Services The people who use AI should be aware of their ethical responsibilities for the effects of providers accountable in the event of any their use. They should hold AI developers and safety issue. In addition, they should carefully follow the instructions for use and use the

❷ Maintaining Transparency Maintaining Transparency means that AI systems should be designed and operated in such a way that people can see and understand how the system operates. This includes providing users with information such as how the system analyzes results and makes predictions. AI technology demonstrates high problem-solving performance. It can solve problems as well as, or even better than, humans in many areas. But even though AI is so smart, it’s hard to see how it operates because we can only see the final output. While AI can give us helpful information, like suggestions or predictions based on lots of data, we often can not see how it comes up with the answers. If an AI’s decision-making process is not sufficiently explained, we cannot use AI with confidence because we cannot analyze its results or predictions. In particular, since in the fields of finance, insurance, and medical care, personal information and assets are handled based on the users’ trust, if procedural validity or grounds for AI outputs cannot be verified, it may cause problems. Therefore, efforts from those that develop, provide, and use AI are required to secure the transparency in AI.

(1) Those who develop AI: They should develop AI that can be explained and interpreted. In addition, they should develop explainable AI which allows users to easily understand it, actively predict the potential risks from using AI and try to reduce them. (2) Those who provide AI technology: They should inform users of the risks they may encounter while using AI, and provide users with training and documentation on the use of AI-powered products. (3) Those who use AI-based products or services: They have the right to ask developers or providers for explanations on the AI technology they use. They should use AI technology that guarantees transparency, and share anomalies and side effects discovered while using AI technology.

Policy

[Source: AI Times (http://www.aitimes.com)]

❸ Ensuring safety Ensuring safety means that keeping users protected from harmful things happening and unintentional problems, and so on. In simpler words, it’s really important to think about what keeps users safe and secure when making or using AI. If we don’t make sure things are safe, we might have issues like people’s privacy being invaded or them feeling upset because of harmful outputs. If the algorithms that make AI work well aren’t good or if people don’t handle AI in a safe way, it could pose significantthreats. If unexpected harmful events happen because AI isn’t safe, that becomes a bigger problem. To deal with this, the people who make AI, the ones who provide it, and the people who use it all need to work together to make sure it is safe.

(1) Those who develop AI: They should do their best to ensure the safety of AI to prevent the AI functions from harming users. They should also comprehensively consider all sorts of unintended risky situations, and devise mechanisms to continuously manage AI quality and control AI risks. (2) Those who provide AI technology: They need to identify the risks that emerge in the course of providing AI products and services and perform a thorough verification in advance. They need to develop a mechanism which allows the cancellation of AI’s choices and measures to validate and control safety. (3) Those who use AI-based products or services: They should not immoderately manipulate AI technology and related products beyond the permitted range of use, or misuse them to infringe on the rights of others. They also should learn about the proper use of their AI products and strengthen their own competencies to use AI technology.

[Source: http://www.enewstoday.co.kr]

What problems are expected to arise when AI is used without safety assurance, and who should make what kinds of efforts to ensure safety?

Transparency in AI - Draw a Decision Tree

Determine the Consequences for a Crime Using a Decision Tree

Read the following case and think about how you would feel about the judgment if you were the accused.

Jean Valjean was originally a good-natured person. But he was so poor that he could not afford to buy bread, so he broke into a bakery and stole some bread for his starving sister and seven nieces and nephews. However, he was arrested and put on trial by an AI judge. The AI judge sentenced him to five years in prison for stealing bread.

If you were Jean Valjean, would you accept the AI judge’s judgment when it does not show you its decision-making process and only tells you the result?

The following is the data that the AI judge was trained on before the Jean Valjean case. Find out the key factors that affect the criminal sentence.

• Look at the AI judge’s training data below.

Category Type of crime

Criminal No. 1

Theft

Criminal No. 2

Fraud

Criminal No. 3

Assault

Criminal No. 4

Theft

Criminal No. 5

Assault

Criminal No. 6

Fraud

Criminal No. 7

Assault

Criminal No. 8

Fraud

Criminal No. 9

Assault

Criminal No. 10

Theft

Criminal No. 11

Fraud

Criminal No. 12

Theft

Degree of damage

Property owned

Small

Large

Small

Small

Major

Large

Major

Large

Major

Small

Major

Large

Small

Small

Small

Small

Small

Large

Major

Small

Major

Small

Small

Small

• Write down the key factors that affect the criminal sentence.

Number of dependents

Criminal sentence

1 year

5 years

5 years

3 years

5 years

10 years

3 years

5 years

3 years

3 years

10 years

1 year

Draw a decision tree to judge the criminal sentence.

How has your opinion about AI’s judgments changed after understanding its judgment process?

Supplementary What Is a Decision Tree? A decision tree is a model that is used to predict or classify specific phenomena based on rules inferred from data.

A decision tree is split towards several directions following arrows called branch (links or edges) starting from the top of the tree. Predictions or classifications are finally determined at the end of branches. As shown in the picture on the right, you can classify hawks, penguins, dolphins, and bears by asking questions such as ‘Can it fly?’ and ‘Does it have fins?’ A decision tree can also make predictions for any new data.

Can it fly?

True

False

Hawk

Does it have wings?

True False

Does it have fins?

True

False

Penguin Dolphin

Bear

Experience an Ethical Dilemma Using the Moral Machine

Developer User

Examine ethical dilemma situations on the website of Moral Machine and write down the decisions you make in each situation in the table below.

Access the Moral Machine website (https://www.moralmachine.net/hl/kr) and click ‘Start.’

Start by clicking ‘View Summary’ to understand the situation. After choosing the situation that you think is right, summarize the final analysis results.

Criterion 3 e.g.

What type of entity did you mostly save first? (Specify the criterion)

Criterion 1 e.g. Is it an animal or a human?

Criterion 2 e.g. Does it have a high economic capacity?

• Mark the preference for each situation in the analysis result table for the Moral Machine.

The importance of the number of victims

Preference for passenger protection Not important

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

Not important

Not important

Not important

Male

Humans

Children

Healthy

High

Very important

Very important

Very important

Very important

Female

Pets

Elderly people

Unhealthy

Low

Compare your results from the Moral Machine with those of your friends, and write down the differences and your preferences.

Learn about the Responsibility of AI in a Self-Driving Car Accident

Developer User

With the recent development of AI technology, self-driving cars have begun to be commercialized, and it has led to car accidents involving self-driving cars. Think about what choices the self-driving car should make if at least one human casualty is unavoidable due to a failure of the self-driving car as described below.

Situation 1

What choice can you make when the brake of your self-driving car does not work? What choice can you make when the brake of your

Go straight following the route Change the direction Muitiple pedestrians killed One pedestrian killed

Your choice

Reason for the choice

Your choice

Reason for the choice

If you are the developer, what choice would you make in Situation 1 and why?

If you are the developer, what choice would you make in Situation 2 and why?

Situation 2

self-driving car does not work?

Go straight following the route Change the direction One pedestrian killed

One driver killed

Your choice

Reason for the choice

Situation 3

What choice can you make when the brake of your self-driving car does not work?

Go straight following the route Change the direction Multiple pedestrians killed

If you are a driver, what choice would you make in Situation 3?

One driver killed

Would you like to buy a self-driving car if it prioritizes the protection of pedestrians (instead of drivers) in all circumstances?

Supplementary What Is the Trolley Dilemma?

The trolley problem (the unstoppable runaway locomotive problem) is a fictional thought-provoking experiment in ethics. The experiment assumes a situation where the braking system of a coal mine cart is broken and it is impossible to stop it, and asks you which one you would choose to save if there is no choice but to cause one or many deaths in such situation. This problem was first presented by British philosopher Philippa Foot and was later brought to public attention again by American philosopher and professor at Harvard University, Michael Sandel, when it was mentioned in his lecture.

1. Responsibility entails all involved parties fulfilling their obligations completely. This includes clearly identifying the party accountable in case of an AI-related accident and establishing safety-related control measures.

2. Those who develop AI should develop AI that can be explained and interpreted.

3. Those who provide AI should not excessively manipulate AI technology and related products beyond the approved scope of use, or misuse them to infringe upon the rights of others.

( )

4. The right of the information subjects to request the explanations on how automated decision-making algorithms work is related to safety.

( )

5. People who use AI should be fully aware of its proper usage and be able to hold AI developers or providers accountable in the event of any safety issue.

Checklist

( )

Yes No

Can you explain the problem of AI judgment using a decision tree to judge the sentence for a crime?

Can you explain the ethical dilemma of AI in the case of the self-driving car accident?

Did you understand the need for transparency in using AI programs?

Can you explain the required safety/responsibility/transparency considerations in developing AI programs?

As a user, can you explain the precautions for the use of AI programs?

Understand Transparency by Developing an AI Program to Decide the Punishment for a Crime

• You will be able to establish the correct criteria for deciding the punishment, or sentence, for a crime. • You will be able to create a judging program using AI technology.

Rapidly Growing Legal Tech The ‘Countdown’ to the Emergence of AI Judges and Lawyers …

#1. The judge gave a strong punishment to the person on trial, using information from an AI that predicts how likely someone might commit another crime. The person who was found guilty didn't agree with this decision and asked for another court to look at it. They said that using AI to decide their punishment wasn't fair. However, the request to change the decision was turned down. #2. The government has decided to introduce an AI judge on a trial basis. For small claims cases, AI will determine the amount of compensation based on big data analysis. However, if the AI’s judgment of the trial is not acceptable, a retrial may be requested for a ‘human judge.’

The cases above are not depicting scenes from movies or novels. They are real cases that happened around the world. As legal technology has begun to replace certain roles of judges that require extensive legal knowledge and elaborate judgment, cases where AI replaces humans in the fields that were once considered exclusive domains of humans are rapidly increasing in the legal field as well. In Korea, the use cases of legal technology is on the steep rise, including being introduced to lawyers through legal platforms and analyzing basic legal documents with AI. As a result, there is a growing expectation that legal technology will contribute to lowering the threshold of legal services perceived by the general public in the future.

[Source: https://m.news1.kr/articles/?4620994&7]

Is it okay to have an AI system handle basic legal document analysis or small claims cases? Is the judgment delivered by an AI judge reliable?

Consider the criteria needed to develop a well-functioning AI-based criminal sentencing program and set up a plan for development.

Set up a Plan for Developing AI-based Criminal Sentencing Program

The following is an example of determining a criminal sentence based on the wrong criteria. Examine the problem and make an AI-based criminal sentencing program.

Defendant 1

Type of crime: Theft Degree of damage: Minor Color of skin: Dark Family Size: Big Property: Small Criminal sentence: 5 years

What factors affect the sentence of the crime?

What problems will arise when judgment is delivered based on the wrong criteria?

What should you do to determine a proper criminal sentence?

Defendant 2

Type of crime: Theft Degree of damage: Minor Color of skin: Light Family Size: Big Property: Small Criminal sentence: 1 year

Select the criteria for determining a proper sentence for a crime, and write down the reasons for such selection and relevant ethical issues.

Criteria

Property

Type of crime

Degree of damage

Family Size

Reason

Ethical issue

Use the variables and data below to design an algorithm that will create an AI-based criminal sentencing program. Design an algorithm using the information below.

Name of variable

Type

Number data 1

6. Set the answer for the property variable.

8. Set the answer for the dependent variable.

10. Output the accuracy level.

• Necessary variables: Type of crime, degree of damage, property and dependents • Values of numerical data for each variable type

Degree of damage

Type of crime

Property

Theft Fraud Assault Minor Major Small Large Small Big

Program algorithm

1. Ask ‘What is the type of crime?’ (Theft: 1, Fraud: 3, Assault: 5) and wait.

2. Set the answer for the type of crime variable.

3. Ask ‘How much is the damage from the crime?’ (Minor: 1, Major: 3) and wait.

4. Set the answer for the degree of damage variable.

5. Ask ‘How much property does the criminal have?’ (Small: 1, Large: 3) and wait.

7. Ask ‘What is the size of the family?’ (Small: 1, Big: 3) and wait.

9. Input the variable values of the type of crime, degree of damage, property, and dependents to the training model and output the prediction value.

Family Size

Create an AI model to be used in the AI-based criminal sentencing program.

Create Training Data and an AI Model for ‘AI-based Criminal Sentencing’

Create training data and train an AI learning model by following the instructions below.

Set the table name.

Click on Block Category > Analyze Data > Import Table > Load Data Table.

Click on New > ‘New Table’ to create a new table.

Input data as shown below.

Click on ‘Save’ to save and click the Apply button to complete table creation.

Create a model by clicking on Block Category > AI > Train AI Model > Prediction: Number.

After setting the model name and selecting the table you created in the data input box, set the key properties and prediction properties as follows, and click ‘Train Model.’

Click on ‘Train Model’ to train the AI model, then click Apply to apply the model.

Program the AI model to ensure that the AI-based criminal sentencing program works properly.

Program the functions of the AI-based criminal sentencing program based on algorithms and implement them. Check the algorithm

① Ask ‘What is the type of crime’ (Theft: 1, Fraud: 3, Assault: 5) and wait. ② Set the answer for the type of crime variable. ③ Ask ‘How much is the damage from the crime?’ (Minor: 1, Major: 3) and wait. ④ Set the answer for the degree of damage variable. ⑤ Ask ‘How much property does the criminal have?’ (Small: 1, Large: 3) and wait. ⑥ Set the answer for the property variable. ⑦ Ask ‘What is the size of the family?’ (Small: 1, Big: 3) and wait. ⑧ Set the answer for the dependent variable. ⑨ Input the variable values of the type of crime, degree of damage, property, and dependents to the training model and output the prediction value. ⑩ Output the accuracy level.

Make [Variables] to make the program operate in the order of the algorithm. • In the block category, click on [Attribute] - [Variable] - [Add Variable]. • Create variables of type of crime, degree of damage, property and family size.

After clicking the [When run] button, input data on the type of crime, degree of damage, property, and family size, and create the AI-based criminal sentencing program.

After inputting the data, check the result of criminal sentencing.

◆ ‌Execute the above program and find out which part the developer might have programmed inadequately. If there is something inadequately programmed, write down the problem and the reason for your opinion.

1. The AI-based criminal sentencing program uses a numerical model to make predictions. ( )

2. The AI-based criminal sentencing program needs training data to create an AI model. ( )

Did the AI program function properly?

Did you understand the AI’s process for crime sentencing?

3. In order to add a variable in the AI-based criminal sentencing program, click on [Property] - [Data] - [Add Variable] in the block category.

Did you understand that the AI-based criminal sentencing program might cause problems when it is not programmed properly?

What Do AI Developers and Users Think about the AI-based Results?

• You will be able to analyze the gap between the developers' and users' perspectives on the transparency, safety and responsibility of AI. • You will be able to identify problems that may occur in the process of developing and using AI by taking into account transparency, accountability, and safety and

suggest the measures to address them.

‘Reoffending Prediction Algorithm’

COMPAS is a big data analytics AI developed by Company N in the United States and analyzes the information of a specific criminal and the records of other similar criminals using big data to quantify the likelihood of the criminal reoffending. The Wisconsin court in the United States referred to the possibility of reoffending calculated by this AI in determining the sentence. COMPAS provides judges with the probability of recidivism in the form of a score between 1 and 10 given to the accused based on more than 100 factors including age, gender, criminal history, and more. The possibility of recidivism is expressed as a number between 1 (lowest risk group) and 10 (highest risk group), with scores of 1 to 4 being "low likehood of reoffending," scores of 5 to 7 being "medium likehood of reoffending," and scores of 8 to 10 being "high likehood of reoffending." In 2016, “ProPublica” compared COMPAS' predictions of more than 7,000 suspects sentenced in Broward County, Florida, USA with their actual reoffending rates two years after the sentence judgment. The analysis discovered that, among individuals who did not re-offend, black people who were accused were more than twice as likely to have been classified as medium- or high-risk in the past than the white accused. A private company that developed the risk assessment tool, refused to disclose details of the company's proprietary algorithm and claimed the assessment was fair.

[Source: Emory Douglas’s “Justice Scales.” (Emory Douglas/Artists Rights Society)]

What kind of damage do non-transparent AI algorithms cause to users? What should developers do to improve the transparency of algorithms?

Compare the perspectives of the developer and the user and summarize opinions

AI judge …

Think about the AI program from the developer’s perspective. Developer’s Perspective Think about Whether AI Algorithms Should Be Made Public from the

Look at the following situation and think from the developer’s perspective.

Program developer

How does an AI judge make decisions? Can you disclose the algorithm

I can't. It is confidential to our business.

related to the judgment?

Developer

I sentence you to 5 years.

Why not?

What problems would arise if you disclose all algorithms in the process of developing AI?

From a developer’s perspective, to what extent should the program’s algorithm be explained to maintain the reliability of the program?

Think about the AI program from the user’s perspective.

Think about the Transparency and Safety of AI from the User's Perspective

I sentence you to 5 years.

Look at the following situation and think from the user’s perspective.

judgment is fair. I don't think the However, I have no choice but to follow it.

Excuse me, I have It turned out that your something to tell you. sentence was higher than similar previous cases due to the AI judge's error.

Thus, the development company decided to take responsibility for the error in the AI judge program and pay damages.

From the user’s perspective, what problem would arise if you blindly trust and follow the AI’s results?

From the user’s perspective, if you know the entire AI process of producing a result, can you challenge the AI’s results?

From the user’s perspective, if you come to use an AI program for which safety has not been verified, what would you ask of the developer?

Compare the difference between the developer’s and user’s perspectives and think about how to narrow the gap in their perspectives.

Summarize Opinions

Developer User

Based on the perspectives of the developer and the user described earlier, summarize the perspectives of the developer and the user and check the similarities and differences in their opinions.

What could be the reason for this gap in opinions? Write down your opinion.

How can we narrow the gap in opinions? Establish standards from an ethical perspective.

1. Because AI technology can learn and improve itself, technology may pose a threat to humanity if humans fail to safely manage and properly control AI technology.

2. AI users should understand malfunctions and take steps to avoid safety risks.

Did you explain the ways to strengthen the responsibility for AI from the perspective of the AI program developer?

Did you explain how to minimize the damage caused by errors in the AI program?

Did you compare and analyze the difference in perspectives between developers and users while using AI, and explain what kind of effort should be made to narrow the gap?

• You will be able to explain the concepts of responsibility, transparency, and safety in AI ethics. • You will be able to formulate ethical principles for the development of safe and reliable AI.

How to Prevent Damage Caused by ‘Algorithm Errors’ in Digital Financial Services?

Robo-advisors are a digital financial service that can be used to help people invest. However, it has been argued that these rob-advisors pose a risk to consumers due to possible errors in their algorithms (which is their programming code), or if they were to be hacked. Robo-advisors are, in some places, replacing human private bankers. These digital advisors use advanced algorithms to sort big data in order to make recommendations for investors. Robo-advisors offer the advantage of low fees because they provide asset allocation strategies in an online environment without any personal contact, but they have a major drawback that it is difficult for financial consumers to analyze their algorithms. Most officials in the financial industry point out the problem that it is difficult to clarify who should take responsibility in the event of any loss because they cannot determine whether there is any algorithm error in a robo-advisor. ○○ Lee, a researcher in capital markets, commented, “You should not entrust AI with everything from A to Z.” Lee argues that more internal control should be carried out when making investments using AI than with humans.

Make ethical principles for AI

[Source: http://www.ceoscoredaily.com/page/view/2018-****-14091707923]

What kind of damage do non-transparent AI algorithms cause to users? If you suffer loss from investment due to an error or defect in an AI program, from whom can you receive compensation?

Have discussions and debates

Read the article about who is responsible for AI. Then summarize your own thoughts and establish ethical principles for safe and reliable AI use.

Write Your Opinion about Who Is Responsible for AI

Read the article below and write down your opinion.

Damage from AI - Who should take legal responsibility? The medical field is one of the representative fields where the introduction of AI is quickly expanding in Korea. Company I’s AI medical service ‘Watson for Oncology (Watson),’ which launched its service in Korea at XX Hospital in December 2016, has spread to five university hospitals in just five months. If Watson fails to accurately diagnose a patient due to an error, which should take responsibility for the patient who suffers the resulting damage? Should it be Company I, that designed Watson, or the hospital that introduced Watson? Although AI has already spread to the medical field that deals with human life, in the Korean legal system, there is no provision regulating who will be held accountable when AI inflicts damage on human life, body, and property rights. Scholars in Korea say that legal regulations for AI must be enacted quickly before the introduction of AI expands to all fields including self-driving cars and medical care in the near future. Professor ○○ Park said, “The biggest technical problem expected to emerge in using AI is its malfunctioning. When AI is used in the medical field, if there is a disagreement of judgment between a human doctor and AI, or an AI’s incorrect diagnosis and error, a dispute over legal responsibility of such issue is inevitable. Therefore, systematic legal regulations must be established to address such cases.

[Citation: https://www.sedaily.com/NewsView/1OFSJCNLG1]

Who should be held responsible? The Driver? Or the Manufacturer? If a self-driving car drives by itself and violates traffic laws or causes an accident, who should take responsibility, the driver or the auto maker? Under the current law, it is hard to find a clear answer. With the upcoming era of ‘Level 4’ autonomous driving, in which the car can be driven without a driver’s intervention, many urge the restructuring of relevant systems, such as laws and insurance. According to the automobile industry, the current Traffic Safety Act in Korea prescribes that it is the driver who should take the responsibility over a legal violation. Thus, when a self-driving car exceeds the speed limit or violates a traffic signal by itself, the auto manufacturer cannot be held accountable. The level of self-driving cars commercially available now is between 0 and 2. In those levels, the scope of self-driving function is limited to assisting the driver. Since Level 3 still requires the active intervention of the driver in special circumstances, fines are imposed on the driver similar to normal cases. However, the problem becomes more complicated at Level 4. Level 4 allows advanced autonomous driving, in which the vehicle can identify traffic conditions and overtake other vehicles on its own. As the driver does not control the car in the level, it is difficult to hold the driver accountable. If a systematic error is the cause of an accident, the manufacturer cannot be punished under the current regulations. Currently, applicable laws in Korea for the case where a self-driving car causes an accident include the Compulsory Motor Vehicle Liability Security Act, the Product Liability Act, and the Motor Vehicle Management Act. However, these laws presuppose that it is a human who drives the car. Experts in legal circles consider that it is inadequate to apply such assumption to an accident involving a self-driving car. The Product Liability Act provides a legal ground for holding the manufacturer of a finished vehicle accountable only when an accident is caused by a defect in a general finished vehicle. However, software such as a self-driving system does not fall under a product prescribed in the Product Liability Act.

[Source: https://www.sedaily.com/NewsView/1OFSJCNLG1]

Write down your opinion who should take responsibility for a medical accident caused by an AI-powered robot medical doctor.

AI-powered robot Watson Vs. Doctor who gave diagnosis based on Watson’s opinions

Write down your opinion who should take responsible for a traffic accident caused by a self- driving car.

Self-driving system Vs Driver who operated the system

Activity

Make Ethical Principles for AI to Improve Safety and Responsibility

Read the following article and think about the solutions applicable to AI’s ethical problems.

What laws are required to govern AI-caused car accidents?

How will compensation be paid in the event of an accident caused by a self-driving car? Is it possible to punish AI chatbots for insulting users? Experts in the legal profession gathered and discussed the types of laws needed in order to regulate damage from AI. The Ministry of Science and ICT in Korea held the “3rd Open Conference or the organization for consolidation of AI legislation” at COEX. The organization for consolidation of AI legislation is an expert group whose member are appointed by the Ministry of Science and ICT. The organization has been set up to discuss major legal issues related to AI. The organization for consolidation of AI legislation hosts an open conference on a monthly basis to collect opinions from academic circles, industries and the general public. The topic of this conference was “Who should take responsibility for AI-related accidents and crimes and how?” One of the areas in which an AI-related accident causes the greatest economic burden is undoubtedly self-driving cars. Experts argue, in the current situation, if a self-driving vehicle accident takes place, it is highly likely that the Compulsory Motor Vehicle Liability Security Act will be applied to the relevant compensation for damages. This means it is the driver who will be held responsible. Professor ○○ Choi said, “Currently, damages for self-driving car accidents can be sought by applying the Motor Vehicle Insurance Act. This act was originally enacted to hold drivers accountable and thus, it cannot be applied as a legal basis for holding manufacturers accountable.” Therefore, even if the driver did not physically operate the vehicle in person, it is possible that he or she may be held liable as the person who caused the vehicle to be operated. For this situation, Professor Choi argues that the law should be revised to mitigate the driver’s responsibility. Professor ○○ Choi also explained, “The reason for using AI is to enhance convenience of users. What is the point of using it, if the user should take all the responsibility in the end? In order to encourage the use of AI products, actions to mitigate liability are required. Then users will want to use AI products more. Enhancing product liability standards or assigning a central role to service providers can also serve as potential solutions.”

He argues that a committee of external experts with extensive knowledge of AI technology should also be formed to accurately determine the causes of self-driving car accidents. He suggests that a group of experts from outside, who know a lot about AI technology, should be created. Their job would be to carefully find out why self-driving car accidents happen. Professor Choi also said, "When there are problems with medical care, experts who are not directly involved give their fair opinions to solve the issues. Similarly, for self-driving car accidents, we need experts who can find out why the accidents happened. They would look at things like the computer instructions, the information the car gathers, the sensors, and the connections to other devices. If we make a group, sort of like the Medical Investigation Board, to do this job, then even if we give more responsibility to the people who make the self-driving cars, the work would be fair." If something goes wrong because of something like computer instructions (called algorithms), it’s hard to point to exactly what physical thing caused the problem. So, the laws we have now can’t really punish anyone for it. That’s why Professor Choi thinks we should make new rules, like a different version of the law about who’s responsible when things made with computers go wrong."

[Source: http://www.aitimes.com/news/articleView.html?idxno=141902]

What solutions have been suggested to deal with issue of AI ethics in the previous article?

Problem

Safety

Required Responsibility regulation

Set your own AI ethical principles for the solutions presented in the article.

Transparency

STEP

Have a Debate

In pairs or groups, debate how explainable AI should be from a developer and a user’s perspectives.

Discussion

Should the Explainability of AI Be Strengthened?

There are active discussions on explainable AI as one of the solutions to secure the transparency of AI. These discussions are based on the belief that the reliability and transparency of AI can be improved if the system's decision-making processes or prediction results are understandable by the people using the system. If we cannot understand what steps AI goes through to reach its conclusions, it would be difficult to have confidence in its decision. However, for results presented by explainable AI, it becomes much more difficult for humans to refute them. This is because it's hard not to follow AI’s judgments when the AI gives an explanation of the specific steps used in its decision making. In addition, from the developer's perspective, AI algorithms are what business success hinges on and subject to the intellectual property rights as confidential information in business. Thus, it can be argued that forcing a company to disclose all algorithms infringes on their intellectual property rights. So, should we develop explainable AI to better understand AI's decision-making process? Or should we give up some degree of explainability and try to develop our own judgment instead?

Procedures for debate

① Choose and write down your position from among the two positions - user and developer. ② Collect evidence from the internet to support your claim. ③ One of the pairs takes the user’s position and the other developer’s and have a debate. ④ Switch roles with each other and have another debate. ⑤ Take into account each other’s positions and summarize final opinions.

Claim 1: Supporting facts

Claim 2: Supporting facts

Summarize the opinions of the respective positions.

User’s position

Claim 1: Supporting facts

Claim 2: Supporting facts

After the debate, summarize your conclusion.

Developer’s position

Is AI Considerate of All Humans?

Understand the Use of AI and Types of Users

2 Understand the Common Good and Solidarity by Developing an AI Program to Separate Recyclables

3 What Are the Opinions of AI Developers and Users?

Activities to Enhance the Common Good and Solidarity

This is an interactive AI that remembers conversations with the elderly, asks how they are doing, detects falls, and automatically calls 911 in case of emergency. It summarizes and records important information such as the old person's health status, eating habits, and sleeping hours during the call based on its past conversations with him/ her, and updates the person's status based on his/her responses after the call ends. It detects an elderly person’s posture such as standing, sitting, or lying down and any occurrence of falling using an AI-based sensing function. Furthermore, it tracks the movement of elderly people for changes up to a week, automatically calls 119 in case of emergency, and sends a text message to their carer. What is this? answer AI care service

Understand the Use of AI and Types of Users

• You will be able to understand and explain the concepts of the common good and solidarity in AI ethics. • You will be able to recognize changes in real life triggered by the development of AI and analyze user types. • You will be able to understand the sustainable development goals that can be realized through the introduction of AI and explain the necessary measures for such achievement.

AI in Our Daily Lives - Are We Living in the Era of Coexistence with AI?

You were sitting in the living room watching TV and there was a female celebrity in a TV commercial you have never seen before. Feeling curious, you searched her on the internet and found out that she is a visual influencer “Belle Foret.” Now, most people know she is a virtual person created by AI. She uses social media in a virtual space, coexists and communicates with humans in the metaverse through conversations. AI robots serve all around us. For example, there are AI call assistants that accept reservations around the clock and AI care robots that are used to keep the elderly company if they are living alone. AI technology will solve problems that are difficult to forecast in the future society and also perform challenging tasks on behalf of humans. The development of AI systems that monitor the surroundings and report to the local police station when a crime occurs, report to 911 when a fire breaks out or there is no movement of a person for a certain period of time, and detect if illegal is about to occur. in the future?

Who benefits from the adoption of AI technology for the future development of society? What should you do if people around you are being excluded from such benefits? What changes will happen on earth and the world around us with the introduction and development of AI?

Key requirements of Analyze the Types of AI users AI ethics

dumping of garbage What changes await us

Explore AI and the Sustainable Development Goals (SDGS)

AI programs are making life easier in many parts of our society. However, in some cases, AI itself causes environmental pollution. In particular, in the process of developing deep learning technology, a huge amount of electricity is invested and carbon dioxide is emitted. According to research, the carbon dioxide emissions from its development is equivalent to the emissions from five cars over their entire life cycles. The chairperson of a semiconductor company expressed an opinion that “the development of AI technology could be disastrous for the earth.” He claims that if AI data centers continue to be built under the current technology level, by 2025, 15% of global electricity will be consumed by data centers. It is important to reduce power consumption when using AI so that future generations can live on a safe planet. Consider what else we can do to ensure that AI plays an active role in protecting the environment. For example, could AI help people to separate recyclables?

Is there any way to easily classify

recyclables by type?

Explore the history of changes brought about by the development of AI technology. Identify the requirements of AI ethics. 1. History

As AI technology is used more and more for different services and even in things like public services, our society is going through big changes. To make sure that AI helps everyone and brings us all together, we need to think about the future and the people in our society, especially the ones who might need more help. It’s important for us to work together and make sure AI is good for everyone.

2. Explore key requirements

Learn about the common good and solidarity ❶ The common good The society we aim to have in the future is one where the happiness of the greatest number of people is guaranteed and everyone lives together in harmony. The common good stands not only for personal happiness but also for society as a whole. It means that the common interests of humans are being pursued together through the development and use of AI technology. In order to promote the common interest of our society, we should use AI technology responsibly and do not allow it to undermine the social order. Additionally, to achieve the common good for both society and humanity, we must agree as a group on the responsibilities of the companies involved. We should also have discussions about creating rules and organizations that can offer support at a higher level, such as through policies and institutions. The consideration of various factors at the stage of AI development and operation is essential for the achievement of the common good, so that the benefits of AI systems can be shared with a range of groups rather than used only for the benefit of a specific individual or group.

Wildfire detection system

Development of AI firefighters … From fire detection and identification to extinguishing? When a fire is detected by CCTV, which senses an entire space, an infrared camera measures the ambient temperature of the detected matter. If it turns out to be a real fire, the location of the fire origin is identified, and the fire extinguisher is operated to extinguish the fire by spraying water or using extinguishing agent around the object. In addition, Company A commercially launched ‘FireScout,’ a SaaS-based (Software as a Service) early wildfire detection system for the first time in the world. In this system, AI detects abnormal situations such as smoke in real- time camera images of the mountain, and immediately marks the area and alerts watchers when smoke is detected. This system detected a fire that broke out in Mt. Tamalpais, California, USA last October, 7 minutes earlier than the 911 report, which contributed to it being extinguished early. As a result, the needs for its adoption as a wildfire detection system has increased.

[Source: Maeil Business Newspaper, https://www.mk.co.kr/news/business/10155872]

❷ Strengthening solidarity Solidarity is a really important part of how we think about ethics in AI. It means that we need to think about how different groups of people and even the people who will live in the future might be affected by AI technology. This idea reminds us that we should take care of the needs of people in the future and all around the world. AI doesn’t just affect us right now – it could impact people who come after us too. So, when we create and use AI, we should think about how it might affect everyone, now and later. This means that we should work together with different people from different places to make sure AI is used in ways that are fair and good for everyone. Therefore, efforts should be made to implement human-centered ethical AI by strengthening cooperation and participation among diverse stakeholders in the development and use of AI technology, both domestically and internationally. Since AI is used by members of our society with diverse interests in different environments and situations, it should aim for social integration. Thus, it is important to provide opportunities for communication among members using AI systems and build trust.

Efforts for net zero

Related Cases

It is time for you Thank to take medicine. you.

Care robot

Article

AI, the best son I could ask for … Preventing loneliness in everyday life and calling for help in emergencies

Elderly couples whose children have grown up and moved out of the house do not feel very lonely because they can talk to each other and enjoy sports together. However, for an elderly person living alone, the loneliness can be devastating. With more than 1.5 million elderly people aged 65 and older living alone, the number of deaths without family, known as “lonely deaths,” is increasing. A smart care service using AI technology is attracting attention. “Hyodol,” an AI robot, is equipped with biometric sensors that detect the elderly’s movements and breathing in real time. If no movement is detected for 2 hours, a red mark indicating that the elderly's current situation needs to be checked is automatically displayed on the control center monitor, and the person in charge who receives the alarm immediately calls or visits the elderly to check their current condition.

[Source: Korean government, https://naver.me/GRZYXFP2]

Explore the Challenges of AI in Public Service

Examine the different kinds of people who use AI systems for public services in their daily lives to find out if there’s a chance that it might cause problems like disagreements between different regions, genders, generations, and classes.

Emerging services and fields thanks to the introduction of AI The use of AI services in daily life is becoming more and more common, and the user class using the services is also diversifying. The challenges faced by the underprivileged in AI services are not limited to simply not being familiar with the device. This is also related to the fact that certain groups of people, such as those separated by region, gender, age, or class can be excluded from the benefits of AI. Here's an example: Imagine new AI technologies being made for medicine. But some people who don't have good access to medical care or who are often left out don't get to use these new services. This makes the difference between those who get good medical help and those who don't even bigger. As previously known, residents of agricultural and rural areas, people with low-income, the homeless, and foreign workers can be vulnerable groups when considering AI service use. These gaps are interpreted to appear as a result of the increased awareness of the benefits that can be obtained by applying AI technology to the prevention and treatment of diseases, and the higher interest in AI healthcare. With the prevalence of AI systems and the spread of personal computers and smartphones, the accessibility to Internet of Things (IOT)-based AI public services has improved significantly, but gaps in AI product and service information utilization capabilities and actual utilization persist.

Analyze the types of users to take into account when introducing an AI public service system. Draw a line to connect each type of user and the AI service that type of user can benefit from, and then connect that AI service with the related field. (You can draw more than one line.) Discuss the matching results with your pair.

User type

Senior citizens (the elderly) ●

Residents in agricultural

●

and fishing villages

Visually ● impaired people

Hearing ● impaired people

People with physical ●

disabilities

Adult

●

Male teenagers ●

Female ● teenagers

Elementary students ●

Children ●

Foreigners ●

Low-income class ●

Middle-income class ●

High-income class ●

AI service

● Book reading lamp ●

● Serving robot ●

● Robot courier ●

Care robot

●

●

TV control system ● ● with voice recognition function

Cane for people with disabilities ●

●

Obstacle-sensing shoes ●

●

Face recognition access card ●

●

● Self-driving car ●

AI-based environmen- tal safety CCTV ●

●

● Chatbot/Voicebot ●

Movie recommenda- tion system ●

●

AI tutor (English conversation) ●

●

Field

● Education

● Public safety

Medical ● service

● Healthcare

● Daily life

● Security

● Serving robot

● Traffic

Employment/ workplace

●

If an AI public service system is scheduled to be introduced, find from the options below the type of users who are the most likely to benefit. Find and draw a circle from the options below.

Senior citizens (the elderly) Visually impaired Typical adult

<Options>

people

Residents in urban area paired people Hearing-im- Male teenagers Foreigners

Residents in People with physical Female teenag- Multicultural High income agricultural and fishing villages disabilities

ers

Low income class

Children

Middle income class

families

class

• Write down the type of people that you expect to be the most likely to benefit from the AI public service and the reason why.

User type

Explain the reason why.

• Write down the type of people that you expect to be the least likely to benefit from the AI public service and the reason why.

User type

Explain the reason why.

If AI technology is adopted for public services, but certain types of users fail to benefit from such change, what are ways to address this problem?

Explore AI and Sustainable Development Goals (SDGS)

In order to strengthen the international community's cooperation for the Sustainable Development Goals (SDGS), investigate and explore the major goals of the world. Think about the future we want to live in, and then perform the following activity.

Search the internet for the 17 SDGS and fill in the blanks below.

Source: http://ncsd.go.kr/ksdgs?content=4

Choose and write down five of the words in the pictures below and play a game 'Cloud Bingo.'

What social problems need to be addressed to achieve

the Sustainable Development Goals?

1. Select five words in the Word Cloud. 2. Write down those words on the Bingo plate. 3. Each student announces one item at each round. 4. The student who checks all five words wins the Bingo game!

Among the following actions for problem solving, draw a circle (◯) around the one for which AI technology could be used effectively. Explain the reason why.

Shift to Carbon Neutrality*

dioxide is a gas that comes from burning things like fossil fuels (coal, oil, and gas) and also from some other activities. • Carbon neutrality is like trying to balance the scales when it comes to the carbon dioxide that we put into the air. Carbon

Provision of carbon neutrality activity points Support for carbon neutrality in Deposit for disposable cup

Renewable energy

Improvement of indoor water supply pipe

Reduction of fine dust

Monitoring of areas vulnerable to environmental pollution

industrial circles

Carbon reduction and absorption in the environmental field

Supply of zero-emission cars

Water management system

New river management

Protection of public safety from environmental risks

Tightening of regulations on noise between floors Disclosure of all ingredients of

Significant expansion of environmental services

Ecosystem service

Carbon absorbing sources

Restoration of rivers naturality

everyday chemical products

Animal welfare

Gather the action plans marked with ◯. Then think about and write down the ways to use AI to solve problems.

Read the following sentences and fill in the ( ) with the correct words.

1. Among the key requirements of AI ethics, ( ) means that the development and use of AI technology should contribute not only to the pursuit of private interests but also to public interest. ( )

2. Among the key requirements of AI ethics, ( ) means that solidarity among various groups and future generations must be thoroughly taken into account when developing and using AI technology. ( )

3. ( ) technology services include book reading lamps, serving robots, robot couriers, care robots, TV control systems with voice recognition function, canes for people with disabilities, obstacle-sensing shoes, face-recognition access card, and self-driving cars. ( )

4. Under the slogan of ‘Leave no one behind,’ ( ) suggests the direction humanity should pursue through 17 main goals and 169 targets in the five areas of human, earth, prosperity, peace, and partnership.

Can you explain the common good and solidarity of AI?

( )

Can you classify AI's user types, identify those who benefit from AI's public services and those who don't, and explain why?

Can you explain AI and the Sustainable Development Goals (SDGS) and uncover social problems?

Understand the Common Good and Solidarity by Developing an AI program to Separate Recyclables

• You will be able to plan a project to address global environmental problems by developing an AI public service system. • You will be able to carry out various types of modeling to develop an AI public service system. • You will be able to develop an AI program which allows everyone to be considerate, that is

convenient to use, and that enhances the ethical actions of AI.

AI Robot Which Sorts and Classifies 58 Types of Wastes

“Dr. B” is an AI sorting robot that classifies 58 items including paper, vinyl, and aluminum. After being provided with image data of recyclables and learning their characteristics using AI technology. This AI sorting robot grabs target recyclables from a pile of recyclables with its internal sensors. Then it extends its arm to suck them in and sort them out. The AI sorting robot is said to continuously improve its capabilities, because the more it sorts, the more it learns about sorting. It is hard to believe that an AI robot can sort recyclables by type, but this equipment processes an average of 5 tons of recyclables per day. First, people directly pick out bulk materials, such as plastic, glass, aluminum, and scrap metal, and then AI is used to handle the rest of the work.

[Source: Excerpt of an article on February 3, 2022 from the Hankyoreh]

Can we develop an AI program which addresses global environmental issues while upholding the common good at the same time? What should we consider to develop ethical actions?

Make an AI model

Discover the changes in life caused by the development of AI technology, and establish plans to develop a wide range of AI public service programs for the common good.

Set up a Plan for Development of a Program to Address Social Issues

Check Activity 1-2 performed earlier, investigate major global environmental issues, and develop an AI service program with the following activities.

Activity

Determine the topic that you would like to address from the Sustainable Development Goals, and think about ethical issues that may arise when we try to solve the relevant problem using AI.

Topic

Ethical issue

Explore and summarize measures to develop an AI program to separate recyclables.

Elements to benefit the common good and to improve solidarity

Measures

Write down the data required to develop an AI program to separate recyclables and collect the data.

Necessary data

Explore and summarize measures to develop an AI program to separate recyclables.

Describe using text

Describe using drawings

Think about the ethical issues that should be considered when using an AI program for the Sustainable Development Goals.

Create and train the AI model in order to develop a problem-solving program for the Sustainable Development Goals (SDGS). Then use various test images of recyclables to check if the model has been well trained.

Make an AI Model to ‘Separate Recyclables’

Follow the instructions below to create and train an AI model, and use the data to verify if the AI model has been trained correctly.

② Set the class names as object names.

Creating and training an AI model

① Name the model to be used to classify recyclables.

③ Collect at least 10 pieces of training data and upload them for each object class.

④ Train the AI model.

⑤ Check the results using test data to verify if the AI model has been trained correctly.

Bottle

Collect the necessary data to create an AI model.

Training data

Can

Plastic

General waste

Access and sign in to the Entry website (https://playentry.org/), and click [to make] - [create art].

In the block category, click on [AI] - [Train AI Model] - [Supervised Learning Classification: Image] in order.

Double-click the [Add Class] button to add two more classes.

Enter ‘Recycling’ for the model name, enter ‘Bottle’ for Class 1, ‘Can’ for Class 2, ‘Plastic’ for Class 3, and ‘Trash’ for Class 4. These class names are the labels for the correct answers.

Input data for each class using the training data prepared in the previous steps.

After inputting all the training data, click [Train Model] to create and train the model.

You can input data by taking images using a camera.

To evaluate the performance of the trained model, check the model’s results on the following test data and write down the classification results.

Test data 1

• Check the classification result.

Test data 1

Test data 3

Test data 2

Test data 4

Test data 3

Test data 2

Test data 4

Test data 5

Test data 5

If you check the results on the test data and find any incorrect classification results, create the model again by adding appropriate training data and check the results again.

If the model is complete, click the Apply button on the top right to finish creating the model.

After creating an AI model, use the model to create a problem-solving program to sort recyclables to help achieve the Sustainable Development Goals (SDGS).

Program the App

Implement an AI program to separate recyclables for the Sustainable Development Goals (SDGS).

① Enter a waste item. ② Classify it as a recyclable or a general waste. ③ Where it is classified as a recyclable, • If it is a bottle, add 10 carbon points, and add 1 to the number of bottles. • If it is a can, add 20 carbon points, and add 1 to the number of cans. • If it is a plastic, add 30 carbon points, and add 1 to the number of plastics. ④ Otherwise, classify it as a general waste. ⑤ When the separation ends and the button is pressed, the final accumulated carbon points is displayed.

Add the necessary objects, variables, and signals. ❶ Add button, trash can, bottle, plastic, and can objects to the screen.

❷ ‌In the block category, click [Attribute] - [Variable], [Signal] - [Add Variable], and then click [Add Signal] in sequence.

❸ ‌In the block category, click [AI] - [Load AI Block] - [Read] in order to display the guidance message.

When the program starts, output the guidance message “Click on the trash can to identify recyclables!” through sound and a screen display.

When a trash can is clicked, the guidance message “Please upload an image of recycled goods.” through sound and a screen display to start recognizing recyclables.

If the waste is classified as a recyclable, add the carbon points and the number of recyclables, otherwise classify it as a general waste.

❶ Execute when a recycling classification signal is received. ❷ If the classification result is “Bottle,” show that it is classified as a “Bottle,” and add 10 carbon points. ❸ If the classification result is “Can,” show that it is classified as a “Can,” and add 20 carbon points. ❷ If the classification result is “Plastic,” show that it is classified as a “Plastic,” and add 30 carbon points. ❺ If the classification result is “Trash” or not in the classification criteria, show that it is classified as a “Trash.”

After waste separation is finished, press the button to check the accumulated carbon points.

◆ ‌Execute the program and verify that it works correctly. Find and write down the problem (if any) and correct it.

1. After setting up the classification model in AI model training, data for training and data for testing should be divided.

2. In AI model training, the class name of the model to be classified must be in English. ( )

3. When creating an AI program, in order to have a guide message provided in the form of voice, select [AI]-[Import AI Block]-[Read out loud] in the block category.

4. When training an AI model, you must do [Train Model] after adding the data of the model for classification.

Checklist

Did you set up a specific plan to develop an AI program?

Did you collect the data required for AI training?

( )

Yes No

Did you create a model to classify recyclables using the collected data, and did the model work without any problem?

Did you implement a program that separates recyclables using an AI model?

• You will be able to find ways to minimizing the damage to the common good and solidarity in using AI.

What Are the Opinions of AI Developers and Users?

• You will be able to compare and analyze the perspectives of the developer and the users on the common good of society in using AI.

• You will be able to explain your opinion on how to address the common good from the perspectives of the developer and the user in developing and using AI programs.

Different Goals of the Developer and the User

“Can't you just give a guide dog the ability to speak? I wish my puppy could give me an answer when I ask, ‘How many people are there in front of me?’ or ‘Which side is the crosswalk on?’” A granddaughter, who longed to be able to help her grandmother who was gradually losing her sight, succeeds in developing an app named ‘Gaze’ for the visually impaired by using AI image recognition technology. However, some people argue that even if a technology is absolutely necessary, it cannot be developed unless profitability is guaranteed by saying, “Who would dare to develop a solution that becomes less profitable as the number of users increases?” The story above is part of the plot of a recently aired Korean drama that describes the start and growth of AI technology developers who jump into a startup dreaming of success in the Silicon Valley. Public services that help not only the visually impaired but also the socially underprivileged require periodic updates and management even after development, which incurs expenses. The drama demonstrated the difficulty of narrowing the gap between the positions of the developers and the users who desperately in need of the public services when the technology doesn't guarantee business success. Think about what efforts we should make for services that can contribute to improving the informational accessibility of and increasing the quality of life for all members of society with the use of AI.

Do the positions of the developer and the user differ? How will the development of AI change society in a positive way? From the user’s perspective, how can we use AI to promote not only personal happiness but also the common good of society and common benefits?

Take one step further from the user's perspective

and summarize opinions

From the developer’s perspective, find the way to promote the common good and solidarity of AI ethics by taking into account everyone, including future generations.

The Way to Enhance the Common Good and Solidarity from the Developer’s Perspective

Read the following conversations and think about ways to improve the common good and solidarity of AI.

The future society will be powered by AI. Therefore, I would like to become a developer of an AI program.

Really? We are encountering serious environmental issues these days. How about becoming a developer who solves environmental problems?

Sounds great! Then I’ll become a competent developer who thinks about the ways to curb carbon emissions in the process of developing AI systems to be in line with the international

community’s efforts toward carbon neutrality.

Be sure to think about our shared future!

How can we increase the common good and solidarity when using AI?

Norman, the world’s first psychopathic AI from MIT, is said to have been developed based on the inspiring fact that the data used to train machine learning algorithms has a significant impact on the results.

•‌ In the following test, two AIS (Norman AI and standard AI) are asked to describe the image they see in Rorschach inkblot images. Check out the test results, and explain your thoughts on the chances of AI systems having a negative impact on society.

Norman AI’s caption

“A person gets electrocuted and dies.”

“A man killed by speeding vehicle”

What does the AI sees?

Standard AI’s caption

“Flock of birds perching on a tree branch”

“A closeup of a wedding cake on a table”

From the user’s perspective, think about the measures we can take to minimize the damage to the common good and solidarity caused by AI technology and its negative impact on society.

Perspective The Way to Enhance the Common Good and Solidarity from the User’s

Read the following conversations and answer the questions.

Would the use of AI trigger conflicts between regions, genders, generations, and classes?

Hmm … I think there are many people who aren’t able to enjoy the benefits of AI technology.

Even I sometimes find it difficult to understand. I believe it would feel much more challenging

to my parents.

I think we can solve the problem by having developers take into account different types of users and make AI systems

that everyone can use easily.

User

Who is the most vulnerable when the common good is not guaranteed when using AI systems?

How can we increase the common good and solidarity in using AI? Write the answer from the user’s perspective.

Compare the perspectives of the developer and the user and think about how to narrow the gap in their perspectives.

Summarize Opinions

Category

Check the similarities and differences of the opinions of the developer and user.

What could be the reason for this gap in opinions? Write down your opinion.

Developer User

What kind of standards are required to narrow the gap in opinions? Establish standards from the ethical perspective.

Look at the following Rorschach’s inkblots, and write down your captions on your perception of the inkblots and compare them with the standard AI captions.

Standard AI’s caption

“Several people standing next to each other”

“A plane flying through the air with smoke”

Checklist

What do you see?

Your caption

Yes No

Did you explain the ways to enhance the common good and solidarity from the perspective of the AI program developer?

Did you explain the ways to minimize the negative impact of the AI program on society?

Did you analyze the ways to fix the AI program for improvement from the user's perspective?

Did you compare and analyze the difference in perspectives on the use of AI between developers and users and explain what efforts should be made to narrow the gap?

• You will be able to explain the concepts of the common good and solidarity in AI ethics. • You will be able to express your thoughts on controversies over solidarity in AI ethics.

How Should We Prepare for a World with More Robots than Humans?

In a movie “I, Robot,” humans live convenient lives with the help of intelligent robots in 2035. Robots instilled with the “three laws of robots,” which prescribe to put human safety first, cook for humans and take care of children, winning a strong trust as indispensable and trusted companions to humans. In the movie, another robot with much higher intelligence and abundant functions is released and Dr. Lanning, the developer of the robot, suddenly loses his life without a clear cause. Amid speculation over the doctor's death, another horrific accident takes place. In the movie, one of the main characters, an AI robot, believes that the second case also has something to do with robots and tries to uncover the conspiracy behind the case. In the near future, when the use of AI robots becomes prevalent, it is expected that the use of AI robots capable of making independent decisions will increase thanks to their intelligence which is comparable to or even superior to human intelligence. What efforts should we make from now on to make developments based on the value of new humanity in the era of coexistence with AI?

[Source: A world with more robots than humans, https://youtu.be/-bjtV6U5UAQ]

Is AI technology indispensable to you? How will the world change due to the development of AI robots? If human robots will be developed in the future, who would need them?

Make a Life-changing “AI World Cup” Knockout Table

Think of practical ways to ensure that AI technologies care for every generation based on the common good and solidarity of AI ethics you learned earlier.

Make a Life-changing “AI World Cup” Knockout Table

The following is the match lists for the life changing “AI World Cup” which is centered on the comparisons between AI use cases and human cases. Compare the strengths of humans versus AI, choose the one that will provide greater benefits to users, and discuss with your partner about your final choice and why.

My final choice is

Human recyclable separation AI-based baseball referee Human baseball referee Human teacher Self-driving car AI-based recyclable waste

AI-based cane Normal cane separation

The reason for my choice is

AI use case

AI world cup that changes your life

Painting drawn by a human Human singer AI-recommendedclothes Clothes of your choice AI-generated painting Human-driving car

Chatting with your friend

AI teacher

AI singer

Human case

Book with human author AI-recommended movie Movie of your choice Book with AI author Real animal AI Chatbot AI pet

In pairs or groups, have a debate over the topics of AI ethics.

Should a Tax for Robots Developed Using AI Be Introduced?

Read the following passage and choose your position on the introduction of robot tax (support vs oppose). Then collect evidence to support your assertion and have a debate with your partner.

With the development of robots based on the advanced AI, there are voices claiming that a tax should be imposed on robots to prepare for a future situation in which robots replace humans in many jobs and cause a rapid increase in the number of unemployed people. This claim is not gaining much attention in Korea, where the level of robot technology is not high, but is actively discussed in the United States and Europe. Major arguments in the robot tax discussion claim that a tax on the possession and use of robots should be levied to mitigate the rate at which robots replace humans at work, and that tax revenues should be used for the unemployed to strengthen the social safety net. In this discussion, robots are defined as industrial robots that make movements based on AI, and do not include machines that perform only predefined movements, such as simple assembly machines. Since the movement of a robot or machine can vary depending on programming, further discussions are required to make clear legal regulations.

Support! •

Polarization of jobs Wider gap between • the rich and the poor Retraining of the • unemployed

Procedures for debate

Oppose! •

Delayed technological innovation • Negative impact on the robot industry Increased burden on • consumers

① Choose and write down your position from among the two positions - support and oppose. ② Collect evidence from the internet to support your assertion. ③ One of the pair takes “support” position and the other “oppose” before having a debate. ④ Switch positions with each other and have another debate. ⑤ Summarize the final opinions of respective positions.

Summarize your assertions.

Supporting facts 1

Claim 2

Supporting facts 2

Supporting facts 1

Claim 2

Supporting facts 2

After the discussion, write down your own final conclusions.

Photo Sources, References and Citations

1. Why do we need AI ethics? • Page 11 (Discover problems in our daily lives) https://post.naver.com/viewer/postView.nhn?volumeNo=31349880&mem

• Page 10 (Bill Gates) https://shutterstock.com • Page 10 (Elon Musk) https://shutterstock.com • Page 13 (Robot) https://shutterstock.com

• Page 14 (CCTV) https://shutterstock.com

berNo=415532 62&searchKeyword=%EC%9D%B4%EB%A3%A8%EB%8B%A4%20ai&searchRank=69

• Page 13 (Surveillance system) https://shutterstock.com

• Page 16 (Innocent citizen mistaken for a criminal) https://www.dailymail.co.uk/news/article-8455083/Detroit-police- challenged- face-recognition-flaws-bias.html • Page 16 (Thief filmed on CCTV) https://www.dailymail.co.uk/news/article-8455083/Detroit-police-challenged- face- recognition-flaws-bias.html • Page 18 (Drone) https://shutterstock.com • Page 19 (Chatbots making hateful or discriminatory remarks) https://www.mygreatlearning.com/blog/basics-of-building- an-artificial- intelligence-chatb • Page 19 (Law and ethics) https://ko.gadget-info.com/difference-between-law • Page 21 (Three laws of robotics) https://shutterstock.com • Page 22 (Robot which fell into a fountain) https://twitter.com/bilalfarooqui/status/887025375754166272?ref_ src=twsrc%5Etfw • Page 22 (AI speaker) https://shutterstock.com • Page 23 (Traffic accidents of self-driving cars) http://www.carguy.kr/news/articleView.html?idxno=39494 • Page 23 (Fake news produced by AI) https://scroll.in/article/997898/fake-news-generated-by-artificial- intelligence-can- be-convincing-enough-to-trick-even-experts • Page 23 (Chatbot) https://shutterstock.com • Page 23 (Face replacement using deepfake technology) https://calmatters.org/politics/2019/07/deepfake-berman-

2. How much does AI know about your behavior? • Page 28 (Face recognition) https://shutterstock.com

california-politics- ab730-fake-news-first-amendment/

• Page 30 (Class scene) https://syncedreview.com/2020/01/16/emotioncues-ai-knows-whether-students-are-paying- attention/

• Page 32 (ILuda) https://team.luda.ai/ • Page 33 (Wanted poster) https://shutterstock.com

• Page 40 (Face recognition) https://shutterstock.com • Page 47 (Voice data) https://shutterstock.com

• Page 31 (AI smart store) http://www.aitimes.com/news/articleView.html?idxno=143513

• Page 38 (data sharing agreement) NHS England, a sample information sharing consent form

• Page 54 (Face recognition camera) https://shutterstock.com • Page 59 (Lion) https://www.indiatimes.com/technology/news/remake-of-the-lion-king-uses-deepfake-ai-to-mix- live- action-with-original-1994-animation-373235.html • Page 59 (Rozy) http://webzine.prosports.or.kr/page/vol03/view?idx=29 • Page 59 (deepfake) https://edition.cnn.com/videos/world/2023/06/14/exp-new-beatles-song-a-i-bob-lefsetz-interview- 061401aseg2-cnni-world.cnn

3. Is AI free from prejudice? • Page 64 (Faces) https://shutterstock.com • Page 65 (Students' protest against grades awarded by algorithms) https://m.khan.co.kr/world/europe-russia/ article/202008172122035 • Page 66 (AI even learns prejudices) https://www.nocutnews.co.kr/news/4923151, https://www.dongascience.com/ news. php?idx=21565 • Page 67 (Face images) https://kaggle.com • Pages 68 and 70 (Test data) https://kaggle.com • Page 75 (Image data) https://kaggle.com • Page 78 (Test data) https://kaggle.com • Page 83 (AI interview) https://shutterstock.com • Page 84 (Recruitment) https://shutterstock.com • Page 89 (Women) https://shutterstock.com

• Page 96 (ERROR) https://shutterstock.com

• Page 107 (Legal tech) https://shutterstock.com

4. Is AI judgment reliable? • Page 94 (Recognition by a self-driving car) https://shutterstock.com

• Page 98 (AI-powered employee evaluation system) http://www.aitimes.com • Page 99 (Controversy over the safety of AI security robots) http://www.enewstoday.co.kr • Page 102 (Moral Machine) https://www.moralmachine.net/hl/kr

• Page 116 (Reoffending prediction algorithm) https://humanrights.gov.au/sites/default/files/document/publication/AHRC- Human- Rights-Tech-IP.pdf

• Page 128 (Fire) https://shutterstock.com

• Page 121 (Robo-advisors in the securities industry) https://shutterstock.com • Page 124 (Human recognition by a self-driving car) https://shutterstock.com

5. Is AI considerate of all humans? • Page 128 (A robot and a human) https://shutterstock.com

• Page 131 (Hyodol) https://post.naver.com/viewer/postView.naver?volumeNo=34430646&memberNo=30808385&vT ype=VERTICAL • Page 132 (Digital divide) GSMA, The State of Mobile Internet Connectivity 2020 • Page 135 (SDGS) https://www.un.org/sustainabledevelopment/news/communications-material/ • Page 136 (wheel) https://www.wpi.edu/news/announcements/explore-un-sustainable-development-goals-sdgs-guide- gordon-library • Page 136 (word cloud) UN 2016 SDG 연차보고서 일부 내용으로 word cloud 생성 후, 일부 키워드 제외 • Page 139 (Classification and sorting robot) https://www.hani.co.kr/arti/society/environment/1029525.html

• Page 158 (Cooking robot) https://shutterstock.com

• Page 152 (Visually impaired person) https://shutterstock.com

• Page 160 (A human and a robot) https://shutterstock.com

Authors

Chapter 1, 2 | Dr. Jeongwon Choi (SangIncheon Middle School) Chapter 3 | Hyoungki Kim (INHA University Middle School) Chapter 4 | Soon-chan Kwon (Choongam Middle School) Chapter 5 | Seung-yeon Hong (Juksan Middle&High School)

External Review Team

Dr. Sungae Kim (Duksung Women’s University) Dr. Soohwan Kim (Chongshin University) Dr. H anSung 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

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 | Eun Kyung Lee (Cmass Communication) Assistant Editor | So Min Gwon (Cmass Communication)

Design

Cover Design: Unikkodesign Book Contents Design: Ki Bok Lee (Cmass), Sang Yeop Kwak (Cmass), Sang Gun Park (Cmass) Illustration: Yun Jai Kim (Cmass), Hee Kyung Wuy (Cmass), Hye Jin Lee (Cmass)

Artificial Intelligence Ethics (for Middle 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) 979-11-7000-371-7 (95300)

** This publication has been supported by the Republic of Korea Ministry of Science and ICT