From poker to Google to the NYSE

First, they developed an AI that was the first in the world to beat the best professional poker players. Then, their DeepStack project was acquired, along with the implementation team, by Google. "At that point, we realised we could accomplish really big things, so we thought about applying our experience in creating self-learning game algorithms to algorithmic stock trading," says EquiLibre head Martin Schmid. His dream is to run the best trading firm from Prague.

While studying at the Faculty of Mathematics and Physics, you and your classmate (and now long-time colleague) Matej Moravčík were already dedicated to AI – around 2015, when it was far from widely known. You created an AI focused on poker back then. Was it meant to play and win online for you?

Yes, that was exactly our goal (laughs). In our first or second year of our Bachelor's, Matej and I decided we would try to create programmes to play poker for us online.

I assume you quickly became wary of gaming platform administrators. How much did you manage to win before someone caught on?

I’m not sure, it wasn’t a lot by today’s standards, but we earned more than we did at our jobs. At the time, we didn’t have a tool that could beat the best players. Our programme played more or less like an average player, winning against weaker ones here and there. The advantage for us was that we didn’t have to waste time playing; the programme did it for us. We’d just supervise the programme, sometimes they’d win, sometimes they wouldn’t... But gradually, we became more interested in the scientific side of gaming. After our Master’s degrees, we moved on to PhDs, where we worked on developing artificial intelligence, eventually creating a tool that beat the best human poker players.

How did you achieve that?

We were interning at the University of Alberta in Edmonton, Canada, at the time, and it was there that we perfected our DeepStack project.

As a technological milestone, your AI made the cover of Science magazine in 2017, and your DeepStack project was later acquired by Google DeepMind. What was it like to become part of such a corporation?

It was absolutely amazing. DeepMind is essentially a research arm of Google, and the five years we spent there were incredibly rewarding.

What prompted your decision to become completely independent and move into stock trading?

We’ve always set ourselves big challenges. First, we wanted to beat the best poker players. We built a reputation and contacts by doing that. We proved to ourselves that we could accomplish great things. After five years at Google, we thought about how we could apply our experience to something else. That’s when we came up with the idea of building the best trading company in Prague. We leveraged our expertise in creating self-learning algorithms and realised they could be highly effective in markets. If we succeeded, it could make a huge impact (laughs). It was a real intellectual challenge for us.

How does it actually work?

Algorithmic trading is quite common. The difference in our case is that we’re using slightly different algorithms—ones we used in gaming.

What’s your goal?

To trade our algorithms on global exchanges. Today, exchanges aren’t a bunch of people shouting over papers. Instead, it’s all done through servers, where each trading firm runs a program to automate their buying and selling.

Do you have a computer on one of the exchanges?

Yes, at the New York Stock Exchange. It runs a neural network trained with algorithms similar to those used for poker. The main difference is that there’s far less "noise" or unpredictability in financial markets compared to games. We’ve also managed to bring really smart people to Prague from places like California, the UK, and Amsterdam to build a top-notch team here.

Creating self-learning programs is an ongoing process. What stage are you at now? Are you “sitting” on the New York Stock Exchange, reaping the rewards?

I can’t share all the details, but markets are highly regulated. Unfortunately, you can’t just start a small company and place a neural network on the New York Stock Exchange. However, within our first year and a half of operation, we managed to build the first prototype of our technology. We then presented it to various companies that trade on exchanges, looking for potential partners to deploy our algorithms. We successfully formed a partnership in early 2024.

So, your algorithms are essentially your product, which you offer to companies to help them trade more efficiently on the exchange, is that right?

I’d describe it more as a form of barter than selling a product. We deploy our algorithms through partner companies on the exchange.

What’s in it for the companies?

For example, a percentage of our profits.

You mentioned stock market regulations. Could algorithms dangerously manipulate the market?

That’s why we must implement various mechanisms to meet regulatory requirements. At the same time, it’s important to remember that when algorithms perform well, they can improve the overall functioning of the market. Why does a stock trader make money? Either they got lucky, in which case their profit is one-off, or they managed to identify and eliminate market inefficiencies. The latter is a win-win scenario. There are plenty of companies globally that operate this way and make significant profits, benefiting everyone involved.

At the Department of Applied Mathematics, you teach a course called Algorithms in Modern Game Theory. Do you see this as educating your future colleagues and employees?

I try to ensure my students are smart, publish interesting papers, and secure good internships abroad. And yes, two of my current colleagues were once my students.

Right after this interview, you’re off to teach. What do you gain from working at the faculty?

I love working with young, enthusiastic, and talented people. There’s no shortage of them at the faculty, and they have immense potential. I think it’s valuable to help them and also recharge with some of their energy (laughs).