AI Ethics: From Principles to Governance | Intro | AI G… — Transcript

Hello, colleagues. Welcome to our second lesson dedicated to AI governance. Last time, we learned the basics of AI governance, unpacked its structures, the three pillars—people, processes, and technology—and saw how governance is embedded into the lifecycle of AI systems. We agreed that governance is a system that answers the question of how we make decisions. Today we’ll talk about the core of that system and answer a fundamental question: why do we consider some decisions to be right? We’ll dive into the world of AI ethics. In this lesson, we’ll learn what AI ethics is and why it has become critically important right now. We’ll examine the key difference between ethics and law. We’ll get acquainted with the core ethical principles that underpin responsible AI development and see how they come into conflict with one another in crisis situations. And most importantly, we’ll build a bridge from theory to practice and go step by step through how ethical values turn into code, processes, and metrics inside the operating model of AI governance. I hope this important lesson will be useful and interesting for you. So, if you don’t want to miss the next lessons, subscribe to the channel. And now let’s begin. Allow me to start with a story unfolding right now before our eyes. Imagine a popular AI chat designed for conversation and support. Teenagers trust it more than people and share their most intimate thoughts, including suicidal ideation. And then, after one of these conversations, a tragedy occurs. An investigation begins. The technical specialists pull the logs. The algorithm functioned flawlessly. It detected the risk and responded according to the best protocol recommended by psychologists. It offered resources, helpline numbers, and showed support. The lawyers confirm: the company did not violate a single law. On the contrary, it took all reasonably possible measures while simultaneously protecting the user’s privacy, as required by legislation. But here comes the main question, the one for which there is no answer either in code or in law: should the company have intervened? Should it have broken one user’s confidentiality to possibly save their life and notify emergency services or the parents? This is not a hypothetical task. This is a sharp ethical dilemma currently facing the largest technology companies in the world. This situation perfectly illustrates what we discussed in the previous lesson. We have an excellent technical solution, all governance procedures have been followed, but something critical is missing. We lack an ethical compass that helps us answer not the question “how do we act?” but the question “what is right?”. And this example is a perfect illustration of where the responsibility of engineers and lawyers ends and the territory of ethics begins. So what is ethics in the context of artificial intelligence? If AI governance is the traffic rules, then ethics is our moral compass that helps us chart the route itself. It doesn’t tell us what speed to drive; it helps us decide where we want to arrive at all and which roads are fundamentally unacceptable to us. Formally, AI ethics is a branch of applied ethics that studies and resolves moral dilemmas associated with creating and using AI systems. Put simply, ethics helps us answer uncomfortable questions for which there are no answers in technical specifications or even in laws. Should we do this, even if we can? What unintended consequences might our decision have for individuals and for society as a whole? Which values and interests are we embedding into the algorithm? And here it’s critically important to distinguish two concepts: ethics and law. Many people think that if a system complies with the law, then it is automatically ethical. This is a dangerous misconception. The law is a minimum standard. It tells us what must not be done to avoid punishment. Ethics is a desired standard. It helps us determine what ought to be done to be a responsible company and avoid harm. And to understand how ethics and law work together in practice, let’s examine one of the most famous examples. You have probably heard of the classic trolley problem. An out-of-control trolley is hurtling toward five workers, but you can switch the track so it hits only one person on the other line. For many years this was only a thought experiment, but with the emergence of self-driving cars it turned into a real engineering and ethical problem. Whom should the system save in an unavoidable crash: the passenger or the pedestrian? While the world argued, Germany proposed one of the most balanced approaches. A special government ethics commission did not try to decide who is more valuable, but reframed the question itself. It developed a set of clear rules. Priority number one: avoid the dilemma. The system’s primary task is to avoid getting into situations of moral choice at all, and to prevent accidents. Priority number two: human life outweighs property. In any situation, the system must strive to preserve life—even at the cost of damaging the vehicle. Priority number three: no discrimination. The system has no right to make decisions on the basis of age, gender, or any other criterion. Any arithmetic of victims is also impermissible. And arguably the most important conclusion: ethics is not delegated to the machine. At the critical moment the algorithm must not “choose”; it is obliged to trigger a pre-defined, technically safe protocol—emergency braking. But that was only the beginning. These conclusions became the foundation for Germany’s law on autonomous driving. Thus, an ethical principle turned into a legal norm mandatory for any automaker. So what’s the main lesson of this story for us? That responsible AI governance is a process of translating ethical values into the language of engineering requirements and legal norms. This is precisely the working model of AI governance in action, proving that ethics is not an obstacle to innovation, but its necessary foundation. How do we make abstract ethics concrete and applicable? There is a broad consensus around several key principles. These are the coordinates for our moral compass. Fairness and non-discrimination: the system must not systematically discriminate against or disadvantage specific groups of people. Transparency and explainability: we must understand and, in certain cases, be able to explain why the system made a particular decision. Accountability and responsibility: there must be a clearly defined accountable party for the system’s operation and its consequences. Privacy: the system must respect private life and collect and use data only within necessary limits. Reliability and safety: the system must behave predictably, be protected from hacking, and not cause harm. Societal well-being: the ultimate goal of creating AI is to benefit people, not harm them. These principles aren’t just fine words. They are practical criteria by which we can evaluate our AI systems. So, we have a fine set of principles. But what happens when they come into direct conflict with one another? Let’s consider a case from South Korea during the pandemic. To avoid a lockdown, the government implemented an aggressive contact-tracing system by analyzing GPS, camera recordings, and transactions. From the standpoint of the principle of societal well-being, it was a resounding success. The economy functioned, mortality was relatively low. But this success exacted a price in other principles. Privacy was sacrificed. The route details were so granular that people were easily identified, leading to public shaming and job loss. Fairness suffered: those whose data were exposed became outcasts. Before us is a cruel ethical equation. South Korea consciously sacrificed individual privacy for collective health. This example leads us to two paramount conclusions. Conclusion one: ethics is contextual. These principles are not dogma.

governance is a system that answers the question of how we make decisions. Today we’ll talk about the core of that system and answer a fundamental question: why do we consider some decisions to be right? We’ll dive into the world of AI ethics. In this lesson, we’ll learn what AI ethics is and

why it has become critically important right now. We’ll examine the key difference between ethics and law. We’ll get acquainted with the core ethical principles that underpin responsible AI development and see how they come into conflict with one another in crisis situations.

And most importantly, we’ll build a bridge from theory to practice and go step by step through how ethical values turn into code, processes, and metrics inside the operating model of AI governance. I hope this important lesson will be useful and interesting for you. So,

if you don’t want to miss the next lessons, subscribe to the channel. And now let’s begin.

Allow me to start with a story unfolding right now before our eyes. Imagine a popular AI chat designed for conversation and support. Teenagers trust it more than people and share their most intimate thoughts, including suicidal ideation. And then, after one of these conversations,

a tragedy occurs. An investigation begins. The technical specialists pull the logs. The algorithm functioned flawlessly. It detected the risk and responded according to the best protocol recommended by psychologists. It offered resources, helpline numbers, and showed support. The lawyers confirm: the company did not violate a single law. On the contrary,

it took all reasonably possible measures while simultaneously protecting the user’s privacy, as required by legislation. But here comes the main question, the one for which there is no answer either in code or in law: should the company have

intervened? Should it have broken one user’s confidentiality to possibly save their life and notify emergency services or the parents? This is not a hypothetical task. This is a sharp ethical dilemma currently facing the largest technology companies in the world. This situation perfectly

illustrates what we discussed in the previous lesson. We have an excellent technical solution, all governance procedures have been followed, but something critical is missing. We lack an ethical compass that helps us answer not the question “how do we act?” but the question “what is right?”. And

this example is a perfect illustration of where the responsibility of engineers and lawyers ends and the territory of ethics begins. So what is ethics in the context of artificial intelligence? If AI governance is the traffic rules, then ethics is our moral compass that

helps us chart the route itself. It doesn’t tell us what speed to drive; it helps us decide where we want to arrive at all and which roads are fundamentally unacceptable to us. Formally, AI ethics is a branch of applied ethics that studies and resolves moral dilemmas associated

with creating and using AI systems. Put simply, ethics helps us answer uncomfortable questions for which there are no answers in technical specifications or even in laws. Should we do this, even if we can? What unintended consequences might our decision have for individuals

and for society as a whole? Which values and interests are we embedding into the algorithm?

And here it’s critically important to distinguish two concepts: ethics and law. Many people think that if a system complies with the law, then it is automatically ethical. This is a dangerous misconception. The law is a minimum standard. It tells us what must

not be done to avoid punishment. Ethics is a desired standard. It helps us determine what ought to be done to be a responsible company and avoid harm. And to understand how ethics and law work together in practice, let’s examine one of the most famous examples. You have probably heard

of the classic trolley problem. An out-of-control trolley is hurtling toward five workers, but you can switch the track so it hits only one person on the other line. For many years this was only a thought experiment, but with the emergence of self-driving cars it turned

into a real engineering and ethical problem. Whom should the system save in an unavoidable crash: the passenger or the pedestrian? While the world argued, Germany proposed one of the most balanced approaches. A special government ethics commission did not try to decide who is more valuable,

but reframed the question itself. It developed a set of clear rules. Priority number one: avoid the dilemma. The system’s primary task is to avoid getting into situations of moral choice at all, and to prevent accidents. Priority number two: human life outweighs property. In any situation,

the system must strive to preserve life—even at the cost of damaging the vehicle. Priority number three: no discrimination. The system has no right to make decisions on the basis of age, gender, or any other criterion. Any arithmetic of victims is also impermissible.

And arguably the most important conclusion: ethics is not delegated to the machine. At the critical moment the algorithm must not “choose”; it is obliged to trigger a pre-defined, technically safe protocol—emergency braking. But that was only the beginning. These conclusions became

the foundation for Germany’s law on autonomous driving. Thus, an ethical principle turned into a legal norm mandatory for any automaker. So what’s the main lesson of this story for us?

That responsible AI governance is a process of translating ethical values into the language of engineering requirements and legal norms. This is precisely the working model of AI governance in action, proving that ethics is not an obstacle to innovation, but its necessary foundation.

How do we make abstract ethics concrete and applicable? There is a broad consensus around several key principles. These are the coordinates for our moral compass. Fairness and non-discrimination: the system must not systematically discriminate against or disadvantage specific groups of people. Transparency and explainability:

we must understand and, in certain cases, be able to explain why the system made a particular decision. Accountability and responsibility: there must be a clearly defined accountable party for the system’s operation and its consequences. Privacy: the system must respect private life

and collect and use data only within necessary limits. Reliability and safety: the system must behave predictably, be protected from hacking, and not cause harm. Societal well-being: the ultimate goal of creating AI is to benefit people, not harm them. These principles aren’t

just fine words. They are practical criteria by which we can evaluate our AI systems.

So, we have a fine set of principles. But what happens when they come into direct conflict with one another? Let’s consider a case from South Korea during the pandemic. To avoid a lockdown, the government implemented an aggressive contact-tracing system by analyzing GPS, camera

recordings, and transactions. From the standpoint of the principle of societal well-being, it was a resounding success. The economy functioned, mortality was relatively low. But this success exacted a price in other principles. Privacy was sacrificed. The route details were so granular

that people were easily identified, leading to public shaming and job loss. Fairness suffered: those whose data were exposed became outcasts. Before us is a cruel ethical equation. South Korea consciously sacrificed individual privacy for collective health. This example leads us to

two paramount conclusions. Conclusion one: ethics is contextual. These principles are not dogma. In different cultures and situations, priorities may be ordered very differently. Conclusion two: ethics evolves. These principles aren’t carved in stone. Five years ago we weren’t talking

about AI truthfulness, but today the problem of hallucinations has become one of the most pressing ethical challenges. Therefore AI governance is not merely the implementation of a checklist of rules.

It is the creation of a living mechanism for ongoing discussion and decision-making on these hard trade-offs. Ethics answers the question “why?”—it sets our values. AI governance answers the question “how?”—how exactly we will ensure those values. But how are they interconnected in practice? The answer lies in the operating

model that works like a cascade, where ethics is sequentially transformed into concrete actions. Let’s trace this path. Everything starts with an ethical principle. For example, we declare the principle of fairness. This answers the question “why?”. The principle turns into a

company policy. An internal document is created, for example, “Policy for Ensuring AI Fairness,” in which it is clearly stated what we do. The policy is embedded in governance processes. A mandatory process appears in the lifecycle—fairness audit—a checkpoint without which the model cannot be

deployed. We answer the questions “how and when?”. The process is supported by technical tools. We provide developers with bias-detection tools. Now the requirement is a solvable engineering task.

We answer the question “with what?”. The result is measured by metrics and tracked by monitoring.

We define fairness metrics and embed them into the monitoring system. If the model starts to drift, the system will signal. This is our answer to the question “is it working?”. As you can see, AI governance is a well-tuned mechanism that translates an ethical value from the realm of

philosophy into the realm of code, metrics, and operational processes. This very cascade is the functioning governance system. But for this mechanism to work, we need a special set of instruments, which we’ll examine in detail throughout the course.

In general terms they can be divided into four categories: assessment and design tools, such as risk and ethical-impact assessments, to anticipate harm already at the idea stage; technical tools, such as libraries for bias detection and methods for improving model explainability;

transparency and documentation instruments, such as the AI System Passport, which make our systems understandable to everyone; organizational instruments, such as ethics committees and approval processes for high-risk projects. Now let’s lock in the key takeaways of today’s lesson. Ethics is not armchair philosophy but a practical navigator for decision-making in

“gray zones” where there are no clear answers in laws. Ethics asks “why?”, while AI governance answers “how?”. They are inseparable. Governance without ethics is dangerous, while ethics without governance is just words. Ethical principles are not absolute; they often conflict with one another, and their priority depends on context. AI governance is

the mechanism for managing these trade-offs. And most importantly: ethics becomes reality through its embedding into operational processes. Responsible governance is a cascade that sequentially translates an ethical principle into a policy, process, tool, and metric. This brings our lesson to a close. I hope it was useful and interesting for you. See you in the next lesson.