How is poker harder than chess or Go for AI? The problem is different.
It is much easier to build a defensive poker algorithm that makes a lot of money from average players and is difficult for even an expert to win from, than it is to build even the most basic chess or Go program. Poker is a far simpler game than either of those, with far fewer choices and much simpler resolution.
Expert poker, however, requires an ability to learn from a player that is attempting to mislead you. This makes even the simplest problems difficult.
Consider one of the simplest possible games, ro-sham-bo (or paper-scissors-rock). To avoid losing, just pay each with 1/3 probability, trivial for a computer. But expert players versus humans win much more than half the time. Computers that try to predict human play and exploit it for winning more than half the time easily win over average players, but have great difficulty with experts.
If you simplify the poker game, two players, limited rounds, limited chip stack or limit play, it’s possible to find mathematically optimal solutions that cannot be beaten, but that are not particularly good at extracting advantage from human players. Eventually people will likely find optimal solutions for more complicated games, but I would expect them to have this same feature. A poker player is not considered an expert for never losing, skill in poker is measured by ability to win.
Computers are close to the point, or perhaps already there, when they can win consistently from even expert players. However we don’t know if people will emerge who study the best computer players and learn to win from them.
This question originally appeared on Quora. the place to gain and share knowledge, empowering people to learn from others and better understand the world. You can follow Quora on Twitter, Facebook, and Google+. More questions: