How far can AI go? An interview with Alexander Panov.

— Alexander Igorevich, you are one of the leading specialists in the field of artificial intelligence. How do you assess its role in modern society?

— Artificial intelligence is a complex phenomenon. On one hand, it pertains to the realm of fundamental science. This field is actively evolving, with new approaches and algorithms being proposed and developed continuously. On the other hand, AI-based technologies are now implemented in a vast number of practical applications. For the first time in human history, we are witnessing such rapid and large-scale integration of scientific achievements into various systems and devices for practical use in everyday life. Numerous startups are actively creating new services in this area.

AI has also become a publicly accessible tool for managing machines and processing information. Each of us, often without realizing it, uses these technologies in one form or another—from applications for gadgets to software for nuclear power plants. In complex systems, AI is employed for forecasting and process management. For instance, in the banking sector, these approaches are utilized in scoring—assessing the risk of loan defaults.

— How do you evaluate the level of Russian science against the backdrop of the global scientific community in the field of artificial intelligence?

— In certain areas, our scientists undoubtedly lead the way. Specifically, in the development of efficient algorithms that allow for savings in computational power. Additionally, in the field of reinforcement learning, we have teams that are performing exceptionally well globally. However, let's say these are fields that are not well-known to everyone.

There are other areas where we are not at the forefront but strive to keep up and maintain the necessary level of competence. At the same time, we mainly replicate what has already been done abroad. For example, this is the case with language models, which are the foundation for popular applications like Chat GPT.

Russian science has its unique features. For example, in the field of AI, our approach is interesting because it is closely linked with other disciplines, such as psychology, neurophysiology, and others. This significantly enriches the research.

— Do you consider yourself a Russian scientist or part of the global scientific community?

— I am proud to belong to the Soviet-Russian school of artificial intelligence. My mentor, Gennady Osipov, was a student of Dmitry Alexandrovich Pospelov, the founder of the Soviet scientific school in AI. Around them, a large collective of scientists formed. This school, like others, has its drawbacks and advantages. Nevertheless, it is a well-defined system of scientific views that allows for a deep understanding of problems and the transmission of knowledge and experience across generations.

— What are your strengths and weaknesses?

— Among my weaknesses, impatience can be noted. This may be related to youth. Specifically, there are tedious tasks that one must accept and perform well, even if they seem boring; for someone, they are important, and this must be acknowledged. However, sometimes patience runs short, which can affect my professional activities.

Among my strengths, I would highlight my communication skills. I can organize a team, negotiate with colleagues, develop a research plan, and create a conducive environment for idea generation.

— How did you come to science and to artificial intelligence?

— In childhood, when asked what I wanted to be, the standard answer was an engineer. Therefore, it is likely that my work is related to engineering themes, albeit with a scientific focus. I studied in a physics-oriented class in Novosibirsk Akademgorodok. I was interested in various disciplines. Among my achievements are victories in all-Russian Olympiads in chemistry. Later, I entered Novosibirsk State University, focusing on the automation of physical processes, which involves scientific programming of equipment. Besides my studies, I worked at the Budker Institute of Nuclear Physics, where I engaged in data analysis for particle accelerators.

During this time, I became fascinated with programming and encountered concepts such as "semantics" and "knowledge integration." While exploring these topics, I learned about methods of artificial intelligence. I then coincidentally discovered the existence of the Russian Association of Artificial Intelligence, whose president was Gennady Semyonovich Osipov. I wrote to him before enrolling in the graduate program at the Moscow Institute of Physics and Technology, and he advised me to come to him, which I did. While studying in graduate school, I got acquainted with all the representatives of the Russian school of artificial intelligence and began actively engaging with these issues. I later entered a doctoral program and wrote my dissertation. In general terms, this is how my trajectory as a scientist unfolded.

— What was the first artificial intelligence system you created?

— One of the weaknesses of AI is the areas where decisions must be made based on limited information. Initially, I worked on such problems, specifically focusing on data analysis and machine learning with small datasets using an approach now known as the DSM method. By the way, the developer of the DSM approach is the domestic scientist Viktor Konstantinovich Finn. He named the method after the initials of philosopher John Stuart Mill, whose reasoning method served as the foundation for the automatic generation of hypotheses.

The advantages of this method are that it allows for the automatic formulation of plausible and credible hypotheses without domain knowledge but through logical constructions. This approach is in demand in systems with a limited dataset. Utilizing these methods, I developed algorithms that immediately demonstrated effectiveness in the field of psychology. Overall, such algorithms have significant practical value as they work well with a small amount of information, thereby complementing statistical methods (e.g., neural networks), which are effective with large samples.

— What needs to be done for Russian science in this field to catch up with the leaders?

— The government's strategy for developing AI is generally correct. It is evident that a lot of attention is paid to scientific research in our country, conferences are held, and applied research is funded. At the same time, new educational programs are being launched, and grants are being awarded—both for applied projects and startups, as well as for fundamental work.

I believe that the conditions necessary to achieve significant successes are in place. However, scientific inquiry is a complex and multifaceted process. It depends not only on funding but also on the human factor. To activate this, it is essential to create and strengthen scientific groups where a "critical mass" of knowledge can accumulate. In the end, quantity will turn into quality, allowing for a breakthrough.

— Where will artificial intelligence be most in demand?

— It is difficult to give a definitive answer. Much depends on the specifics of particular industries. It is clear that in the most popular devices, such as smartphones and laptops, they are adapted first and foremost. The more we use such gadgets, the more we need services that enhance their functionality. Therefore, the demand for "smart" applications—voice assistants, translators, photo enhancement programs, and many others—is rapidly growing.

The application of AI in industry and the economy varies across countries. For example, in Russia, Sberbank claims that thanks to AI methods in the financial sector, the profits of companies within its ecosystem have increased by about a third. In Taiwan, these technologies are widely implemented for the automation and robotization of production processes and assembly lines.

— Can AI become a tool for a breakthrough to a new technological paradigm?

— It is possible. AI occupies an important place in what we understand as progress. It influences all spheres and significantly changes the format of work, communication, information searching, and analysis. Additionally, AI helps create new businesses. Undoubtedly, it is a crucial catalyst for the development of modern society. However, it is difficult to say whether it will provide a leap in science and technology or if we will remain within the bounds of steady, gradual development.

Moreover, when discussing the successes of AI, it should be noted that there are many unresolved problems in this field. Perhaps some of them are fundamentally unsolvable, like faster-than-light travel or perpetual motion machines.

— Can you provide examples of possibly unsolvable fundamental problems?

— For example, developments in the field of Artificial General Intelligence (AGI). In this direction, scientists strive to create a software complex that, similar to human consciousness, can self-learn to perform tasks for which it was not initially designed. In other words, we are talking about creating systems that can think, reason, and make independent decisions. Such machines will possess human-level intelligence, and in the future, potentially superintelligence.

However, to create systems that do not merely generate texts or images and entertain users on their phones, it is necessary to link AI models with the real world. Yet, there are no good architectures that would allow for this.

— What is your team currently working on?

— We are working on AI applications related to robotics. Specifically, we are addressing the challenges of adapting unmanned devices to perform actions in unfamiliar external environments. In this regard, we emphasize the so-called Foundation Behavior Models. These are neural network algorithms through which machines are trained on pre-collected data and human examples, and then further trained in the