How MIT used poker to teach uncertainty
Climbing the steepest part of the learning curve
In 2012, MIT launched a new course that would turn out to be extremely popular. The official title was 15.S50, but everybody came to know it as the ‘MIT poker class’.
The course was the brainchild of PhD student Will Ma, who was studying Operations Research. He had played a lot of poker – and won a lot of money – while an undergraduate in Canada, which had made a lot of people curious when he arrived at MIT, including the head of department. 15.S50 was a legitimate MIT class; if students passed, they could get degree credit.
I’ve always liked poker, both as a game and testing ground for wider ideas relating to uncertainty. Although poker has rules and limits, some key information is always concealed. The same problem crops up in many aspects of life. Negotiations, debates, bargaining; they are all incomplete information games. ‘Poker is a perfect microcosm of many situations we encounter in the real world,’ as Jonathan Schaeffer, a pioneer of game playing AI, once told me.
I interviewed Ma about the MIT poker class while researching The Perfect Bet. Perhaps understandably for a university course, the sessions focused on theory rather than gambling with real money. Given the relatively limited time available, Ma tried to focus on the steepest part of the learning curve.
Here are some of the key takeaways when it comes to decision-making under uncertainty:
Sometimes the optimal move feels reckless. To succeed in poker, you must be prepared to occasionally take a big calculated risk, or walk away from a situation you’ve sunk resources into. The best actions are often more extreme than a beginner is comfortable with.
Confidence is about recovering from mistakes. In one class, lecturer Jennifer Shahade pointed out that ‘confidence is an underrated ingredient to success in chess and poker’. In particular, confident players expect to make errors and move on rather than let one mistake affect their game.
Most people lose by doing too much. When starting out in poker, it can be easy to get bored with folding and instead play too many hands, when you should be watching and learning about your opponents without taking unnecessary risks.
Your decisions change how others play against you. Shahade noted that in chess, ‘most good players forget who they are playing and focus on the position’. But in poker, ‘monitoring your own image and that of other players is a huge part of the game’.
Judge decisions by how they were made, not how they turned out. As Ma put it, ‘I think one of the things poker teaches you very well is that you can often make a good decision but not get a good result, or make a bad decision and get a good result.’
Cover image: Michał Parzuchowski
The idea that confidence means recovering from mistakes rather than avoiding them is such a crucial reframe. Most people think confidence comes from being right all the time, but in enviroments with incomplete information thats literally impossible. What MIT nailed here is teaching students to separate proces from outcome, something traditional education almost never does becuase grading systems punish bad outcomes even when the decisionmaking was sound. I wonder if this changes how students approach their actual research later on.
A timely reminder to help us make sense of some of the decisions that are currently being made by leaders in the 'new world order'.