The research group I was a part of around six years ago made an announcement yesterday that they had effectively solved two player limit Texas Hold'em. There is already a ridiculous amount of press to read on the subject, including Scientific American, the BBC, the Guardian, IEEE Spectrum, The Economist, The Verge, the Washington Post, and many many more.
To give you a rough idea of how impressive this is, the strategy takes 11 terabytes of disk to store, and required 900 CPU years of computing power. In order to make the game even remotely solvable and the solution storable, several breakthroughs needed to be made over the course of several years of research. The Computer Poker Research Group (CPRG) has been making progress towards this goal for a long time. The final breakthroughs include an amazing new compression algorithm for this purpose that greatly decreased the storage requirements of the solution, and an algorithmic improvement to the solver that greatly decreased the number of iterations necessary to converge on a solution.
It's important to note a couple of things:
- This solution is actually an approximation of the game theory optimal solution. It is actually exploitable for less than 1 milliblind/100 hands, or in poker terminology, less than 0.05 big bets per hundred hands. This is a very small amount, and it would be extremely difficult to tell this strategy apart from its best response by playing the two strategies against each other. But it is worth noting.
- This is only a solution for heads-up (two player) fixed limit Texas Hold'em. For the time being, other poker variants have not been solved. Other variants, such as ones where there are three or more players or where the betting is pot-limit or no-limit have not been solved. There are many challenges to overcome before these games are solved.
I plan on writing more extensively on the subject in the near future. In the meantime, I'd like to congratulate Mike Bowling, Neil Burch, Mike Johanson, and Oskari Tammelin on an incredible achievement.