Yandex Scientific Award Laureate Aleksei Shpilman on Developing AI in Russia

What made you switch to data analysis and machine learning? 

I was about to finish one of the best schools in Moscow when I realized that I grew tired of mathematics. I was eager to see the results of my work in the flesh, so when my classmates went to study mechanics and mathematics at the Lomonosov Moscow State University, I opted for the Faculty of Bioengineering and Bioinformatics, which was relatively new at the university. What attracted me most was the program’s diverse curriculum: we had physics, chemistry, biology (even botanics), as well as high-level mathematics and programming. Though I was and still am a techie, I had an interest for life sciences, too, so I decided to specialize in computer modeling of intracellular systems and other related fields. 

As I joined Yandex’s School of Data Analysis, I started to teach bioinformatics, as well. The field has multiple applications – plus the company, among other things, ran studies in bioinformatics. Even though I had only two lectures to introduce non-specialists to the field, I believe my course was, nevertheless, well-received, with some students even choosing to switch their specializations. 

I defended my PhD thesis on computer modeling in biological systems in 2021. Although these days I’m mostly involved in other kinds of tasks, I do my best to continue developing ML solutions for biology and medicine, be it drug design, microscope image processing, or anything else. 

How did you get into IT?

‪Andrei Raigorodskii‬ (Russian mathematician and Presidential Prize in Science laureate – Ed.) spotted me when I was a student and invited me to join his group. For a while, I juggled studies and work but then I had to decide whether I wanted to keep working at the university or focus more on my job at Yandex. I chose the latter – that’s how I got into artificial intelligence. Even now, each time we develop a new system, it seems like magic to me. 

It seems you didn’t leave science behind completely – after all, you did win the Ilya Segalovich Award.

Yandex doesn’t nominate its own employees, so I received this award when I was already at JetBrains. My team and I were awarded for our research achievements. We had a highly efficient joint laboratory, and three members of our group excelled in the category Young Researchers. 

Could you tell us more about your research and how you combined your studies with an IT job? 

How I ended up at JetBrains is a rather common story. I met a girl from St. Petersburg and decided to move there. I was determined to focus on my thesis but within a few days after I quit Yandex, I got an attractive job offer from JetBrains. It was a fortunate coincidence that they were looking for a person with an IT background and I was eager to teach and do science. 

I was offered a developer’s salary and a chance to open my own laboratory. At first, I started with machine and deep learning but then I met ‪Daniel Kudenko‬, a reinforcement learning expert from Leibniz University Hannover, who came to our company. I was so fascinated by the field that I decided to change the lab’s specialization. 

And what is reinforcement learning?

As I see it, reinforcement learning is the most magical aspect of machine learning. Some of its brightest representatives are AlphaGo or OpenAI Five (the first AI that beat the Dota 2 world champions). These are, so to say, self-taught agents, often used in gaming. As per rule, they show excellent performances in modeled environments but not everyday life, due to the time and extensive computing resources needed to train them. Yet we, just as the rest of the scientific community, are taking steps in this direction. 

Could you give us some examples of your projects? 

We had a small-scale experiment that resulted in several publications. We were curious to see whether it’s possible to make a neural network understand the idea of social justice. Essentially, we tried to create some sort of AI government. We used the platform Neural MMO to develop a rather simple massively multiplayer online game (MMOG). There, we had eight types of agents (clans or nations), which, just like in any game of this genre, collected resources, grew, and sometimes fought with each other. And above them all we placed another agent, a kind of overseer who punished them for misbehavior.

Its task was to raise the average life expectancy of in-game nations. The solution the neural network came up with was to get rid of seven nations so that the one left could have a better, longer life. With time, altruistic agents started to appear in the game but kept killing themselves. We had to explain mathematically why this solution wasn’t the best one.

It all may sound a bit crazy, but we were interested in studying the mathematics behind social interactions and AI solutions. This knowledge can help us manage multiagent systems and systems of human-machine communication more efficiently.

What are you working on now? 

Right now I work at Gazprom Neft. I was attracted by the prospect of helming new projects at the national scale – that way, I can be involved with a lot of what’s happening in Russia in terms of AI. I am specifically interested in developing AI technology here, in my home country.

But at a glance, it does not seem like petroleum companies have a lot to do with IT.

Actually, Gazprom Neft is one of the more technologically advanced (especially in the petroleum industry) companies that use AI in their internal business processes. Its subsidiary, Gazprom Neft Digital Solutions, employs several thousand specialists, which makes it one of the biggest IT companies in Russia.

You’ve mentioned that you are particularly invested in the development of AI in Russia on the national scale. What solutions can you offer?

My full title is: Head of AI Development Programs. In addition, I am a research director at the Artificial Intelligence in Industry association. What this all means is that it’s my job to bridge the gap between companies and universities. We are intent on using universities’ research capabilities to improve and upgrade the technologies that are already available. For instance, right now we are working with ITMO to develop a system that would optimize the planning process of setting up production facilities in the field. 

Since I’ve been combining teaching and research at universities with working at commercial enterprises for almost my entire career, I know how to facilitate connections between business and science.

See also:

The Future of AI: ITMO Scientists Present Cutting-Edge Projects

ITMO Launches Open-Source Mentorship Program

ITMO Scientists Win the First Blue Sky Research Innovative Scientific Projects Contest

Recently, you became the head of ITMO’s Software Engineering Master’s program. What are the current goals of this program? Are they going to shift in any way, considering the change in circumstances?

Our primary purpose is to give our students the best possible education. As this program was always one of the finest in the country, its alumni are all highly trained specialists, sought after by every company in the industry. Yet there is substantial attrition: out of 40-50 new students each year, no more than 20 manage to graduate successfully. This program is not for everyone, yet those who can withstand its hardships become some of the best specialists out there. 

Of course, we’ve had to slightly adjust our focus: in the past, the program focused primarily on training software developers. While such specialists are still needed at Gazprom Neft, the company now seeks a lot more data engineers in order to build up its infrastructure. This does not mean that our program will stop teaching students how to use compilers.

I believe that creative freedom is a necessity for quality education. We never say: “Learn exactly these things and you will land a job at Gazprom.” Attracting future employees is up to the company, we simply provide access to all of the latest technology. Ultimately, the students themselves choose what they want to learn, based on their personal preferences.

Everyone agrees that the demand for IT specialists is currently at an all-time high. How can we meet that demand and can it be done quickly?

There was always a need for more IT specialists, but now it is felt much more acutely. There is only one solution, and it is to train more specialists. 

I like this analogy, courtesy of the TV series Newsroom, where a new executive producer is told the following about her job: “There’s a hole in the side of the boat. That hole is never going to be fixed and it’s never going away and you can’t get a new boat, this is your boat. What you have to do is bail water out faster than it’s coming in.” The situation on our hands is similar: we need to train more IT specialists than however many are being lost.

That means that education is our top priority. Regardless of what happens, we are going to need qualified specialists. They will not come out of nowhere – the only remaining option is to train them ourselves, so that is something that I will be doing no matter what happens.

What is the best way for young specialists to gain work experience?

I would say there are exciting projects in each field, be it the space industry or accounting. That’s why students shouldn’t shy away from any opportunity to intern, meet different teams, and network while they are still at university. As of now, Gazprom has lots of projects students can try their hand at.