At the IV Congress of Young Scientists held at the "Sirius" University in Sochi, around 7,000 researchers from 60 countries gathered. Among the speakers at this major scientific forum were scientists from T-Bank. Why do banks need fundamental science and their own laboratories? We discussed this with Daniil Gavrilov, head of the T-Bank AI Research laboratory.
How do scientists open doors of opportunities for businesses?
- It is well-known that banks are all about money. Here we learn that T-Bank has established its own scientific laboratory, T-Bank AI Research, and publishes articles at international conferences. Why are banks turning to science? Haven't all the money already been made?
- I believe it is crucial to understand that the shift of large corporations toward science did not begin today. If we recall Yann LeCun, one of the godfathers of artificial intelligence and deep learning, he started working on convolutional networks back at AT&T Bell Laboratories. This was a major telecommunications company engaged in business that seemed very distant from science. The fact is that all large companies, when creating a product, sooner or later face the necessity to invest in research. If you want to achieve a breakthrough in business and make money, you inevitably have to invest in solving long-term, complex, and science-intensive problems. Therefore, big companies, including T-Bank, open their research laboratories and focus primarily on fundamental science.
- How does this help banks make money?
- New technologies open up windows of opportunities that we never suspected. For example, ChatGPT is a large language model that can summarize a large text, extracting its main essence. Then innovators thought: this tool can be attached to search! If previously you had to browse links yourself, spending time searching for the needed information, now the model can quickly study sources, select and analyze data, and do this work for me in seconds. At the stage when this technology was being created, no one even imagined such possibilities. In other words, science for banks is not about immediate returns and payback; the window of opportunity may not open soon. But if you are there, you will be among the first.
Will corporations take the place of universities in science?
- We are used to universities being the source of knowledge. But in the case of artificial intelligence, the balance has shifted toward industry representatives. Are we witnessing the decline of universities and the old model of education? Will science now be conducted by large corporations?
- I repeat, this did not happen yesterday or the day before. I mentioned AT&T Bell Laboratories, where Yann LeCun worked, which dates back to the late 80s and early 90s. In the case of artificial intelligence, it is clear why the assembly center has shifted from universities to large corporations like Google, OpenAI, and Anthropic. Developing AI requires vast resources. Some mundane things are needed, like investing billions of dollars in graphics processors or video cards. In the last decade, the improvement of artificial intelligence has been driven by scaling training datasets and increasing computational power. Such scaling opportunities are also available to corporations. Currently, we have hit a ceiling, but a different, more intelligent solution will be found, and scaling will begin based on that. This is all a matter of resources—what big tech and infotech companies have but universities do not. Another point is that science is moving from universities to corporations largely because they are the final clients for solving complex intellectual tasks. Consequently, companies need specialists who can engage in cutting-edge developments. For this reason, companies have started to prepare and select talent from university levels and are even beginning to recruit smart individuals straight from schools.
- How many years will it take for education provided by universities created by businesses—like the Central University from T-Bank or the Yandex School of Data Analysis (SHAD)—to become superior to the education at the Physics and Technology Institute or the Mechanics and Mathematics Department of Moscow State University?
- I believe that these are already very powerful tools for training personnel. Even eight years ago, SHAD was already a great place to study machine learning. In my opinion, universities established by businesses will soon occupy very strong positions in higher education.
A diploma means nothing, motivation is everything!
- Can you paint a portrait of the ideal scientist at T-Bank that you would like to see—education, work experience, career trajectory, areas of interest? What qualities should the person you are looking for possess?
- We are not very interested in education in the sense of which university is listed on the candidate's resume. We are even indifferent to the candidate's academic degree because practice has shown that it is poorly correlated with how well a person will work and achieve great results. Of course, it is important that the candidate understands the field of knowledge we work in. But in most cases, the basics are sufficient at the level of a bachelor's degree in higher mathematics. And if a person is sufficiently diligent and motivated, they can learn independently. It is fundamentally important for us that a person is highly motivated to engage in research. This is the key factor. If we see that they are interested in getting to the essence of things, if they are curious and hungry for new knowledge, then we will conduct research with them and write scientific articles. Essentially, all scientific work is driven by curiosity. It is, in any case, a way of life. We just get paid for it.
Has the profession of a scientist or researcher become more attractive to young people? Or is it still difficult to attract youth to science?
- I can't speak for all of science, but in the field of AI, we see significant interest. Neural networks are currently the cutting edge of research. Many young people want to get involved in this. The question is that this is quite a specific job and it is not for everyone. Moreover, not everyone is ready to live at a 24/7 pace when engaged in research. But overall, we see that this field has become quite appealing.
Can strong AI be created within a single country?
- Is artificial intelligence something that can be developed within a single country? Or do we need to collaborate with foreign scientists, attend conferences, and immerse ourselves in the melting pot of international science?
- Let's say that one does not exclude the other. Firstly, we need to attend conferences and interact with other researchers, at least to make Russia an attractive country in the eyes of scientists. Our main goal at T-Bank is to publish in the top scientific journals. Our articles regularly appear there, and they are presented at conferences alongside works from Google DeepMind, Stanford, and other leading global scientific centers. This means we stand shoulder to shoulder with them. What does this mean for foreign scientists? That they can come to Russia and engage in world-class science. This is very important. Can AI be built within the confines of a single country? It all boils down to the resources needed to create artificial intelligence. Suppose someone comes and says: I want a model on the level of OpenAI. The answer is simple: you need investment at the level of OpenAI. Russia has a lot of super talented individuals. To begin with, let's note that the head of science at OpenAI was our compatriot Ilya Sutskever, who was born in Nizhny Novgorod. This year he left to establish a startup called Safe Superintelligence, which will also work on artificial intelligence. So, there are many intellectual resources in Russia; I know this very well because I work in this field. To further accelerate the development of AI, material resources are needed.
- If you were to act as a visionary, what will the future of artificial intelligence look like? What new functions can AI take on?
- The main goal we strive for is for artificial intelligence to learn to solve problems that humans cannot solve. Because currently, even the largest models can only address problems that are common and well-studied. For instance, when I write text in English, which is not my native language, I can easily edit it using models, and it turns out well and is super helpful. But this is a routine task. There are scientific questions whose solutions we do not yet know. For example, the Millennium Problems (which are seven of the most important mathematical problems, with a reward of $1 million for their solution - Ed). The task of proving the Riemann Hypothesis or solving the Navier-Stokes equations using artificial intelligence looks very enticing. This is probably the main frontier in science today.