The First AI Ethics Textbook Is Published in Russia
Ten years ago, it was still widely assumed that only humans could write poetry, compose music and code. Today, artificial intelligence has become firmly embedded in everyday life. AI calculates, writes, develops programs, helps people with daily tasks and even supports creative work. Before anyone had time to blink, AI began taking part in decisions that affect almost every area of life. So, simply being confident that an algorithm works well is no longer enough. It is now equally important to understand what rules it follows and what happens when a system encounters a situation for which its developers did not provide a clear answer.
The National Research Nuclear University MEPhI has published a textbook titled Etika iskusstvennogo intellekta (Ethics of Artificial Intelligence). According to the online media outlet SNG Segodnya (CIS Today), the book examines these issues from two perspectives at once – philosophical and engineering. The authors are Ekaterina Tikhomirova, a professor in MEPhI's Department of Philosophy, Ontology and Theory of Knowledge, and Roman Dushkin, a senior lecturer in the Department of Cybernetics.
From Philosophy to Code
A discussion of AI ethics can easily be narrowed down to an abstract debate about good and evil. But when a real system is being developed, moral principles have to be translated into specific rules, restrictions and procedures. This is where two seemingly incompatible fields, philosophy and programming, meet.
The textbook's authors distinguish among ethics, morality, moral conduct and law. If ethics is treated simply as a set of prohibitions, AI development risks being reduced to a basic rule: anything not prohibited by law is allowed. For technology, however, that is not enough. The law cannot anticipate every situation a complex AI system may encounter. Developers therefore have to consider not only what the system is forbidden to do, but also how it should behave when the right course of action is ambiguous.
Culture and AI
Another issue examined by the textbook is culture. Ideas about fairness, behavior and responsibility can differ from one country to another because of national and regional characteristics. The textbook therefore looks at how questions of morality and behavioral norms in AI use are understood within Chinese, Japanese, Russian and Western traditions.
For people who work with AI, this has practical as well as ethical implications. A system designed for users in different countries encounters not only different languages but also different social norms. Something perceived as neutral or acceptable in one culture may provoke a very different reaction in another and alienate the user. This is particularly important for models that interact directly with people, generate text or images, or are used in sensitive fields such as psychology.
A Code of Conduct for AI
The book examines tools such as RLHF, Constitutional AI, red teaming and governance pipelines. Behind these complex terms are specific tasks: training a model to follow defined human preferences, giving it a set of behavioral principles, deliberately searching for vulnerabilities and unwanted scenarios, and then establishing continuous oversight of the system.
In other words, a developer-defined code of conduct will not by itself make a model follow the rules it is supposed to follow. Those rules have to be built into the model's development, training, testing and operational processes.
Who Is the Textbook For?
The new textbook is aimed at a broad range of professionals. They include students and graduate students in IT fields, AI developers, legal professionals, researchers and people working at the intersection of technology and the humanities.
For a developer, ethical principles can help identify potential product constraints in advance. For an engineer, they can help account for possible failures and ways a system could be misused. For a lawyer, they can clarify why laws and regulations alone may not be enough to govern complex AI systems. For a researcher, they can help assess the consequences that AI use could produce.
The textbook's foreword was written by Russian Academy of Sciences Academician Sergey Glazyev, while the afterword was written by Alexander Ageev, CEO of the Moscow Research Institute of Industrial Policy and Management. The book could serve as a foundation for training professionals whose task is to make intelligent systems powerful, convenient, predictable and safe.
