The Face Behind the Cards: Meet Alex Zhang, President of Pokerbots
Meet Alex, this year’s president of Pokerbots
This interview has been edited for length and clarity.
Alex Zhang ’26 is the president of the recently finished and wildly popular IAP course 6.9630, Pokerbots. Alex sat down with The Tech to discuss training computers to play poker, exciting poker variations, and the rewarding process of organizing such a large event.
So what exactly is Pokerbots?
Pokerbots is a yearly-offered IAP course in which students form teams and train computers to play poker. The first two weeks consist of lectures teaching basic game theory, poker theory, and counterfactual regret minimization strategies that teams can adapt and implement into Machine Learning (ML) models. During this time, students refine their bots by playing them against other teams’ bots on the scrimmage server and even against themselves on local playgrounds. Pokerbots culminates in a final tournament with over $40,000 in prizes and networking with sponsors.
What are some common student challenges?
The biggest challenge for students is not a lack of prior poker background — Zhang himself only started playing seriously once he got to MIT. Instead, it turns out that most students struggled with the limitations set by Pokerbots’ homemade engine system. That is, Pokerbots utilizes a different poker variation each year, so students can’t count on using some bot written in the past; they have to start from scratch. Students have to be able to play 1000 hands in thirty seconds, so it isn’t feasible to just copy a bot found online.
What are some innovative or surprising strategies that have emerged in past competitions?
A few years ago, a team developed a bot that went all-in on every single hand. This bot was surprisingly hard to counter in real-time without an accurate history of its playing style, but since then we’ve mentioned it in lectures and bots with similar strategies don’t do as well anymore. The key insight here is that even given a hand with a high expectation of equity, if it also has high variance because of future cards (the draw and the river), it typically outperforms its equity. Typically, bots that exploit opponent weaknesses, especially those that utilize the most poker theory and heuristics, fare the best.
What’s been the most rewarding part about being Pokerbots president?
It takes a long time to put all this together and we spent nearly all of Christmas break working on it. The best part was just seeing it all come together. We also have a very small team compared to other IAP classes with only nine of us, so it was great to see them all succeed. This year, we had the most sponsors we’ve ever had and also the largest turnout for the final event, so it was very exciting. There were a lot of firsts this year; for example, there was the implementation of a local playground feature and a physical poker tournament without the bots. I think Pokerbots has a very promising future and I’m excited to see where it goes.