Narrow-domain AI has successfully outperformed human players in tic-tac-toe, checkers, backgammon, and chess. Now an AI system developed by researchers at the University of Alberta can defeat poker players.
It’s an interesting development, because this is the first time a machine can defeat the best human players at a game that doesn’t have perfect information. It paves the way for advancements in decision systems that operate on poor information, which is how most of us operate in the real world.
The key was to train a learning algorithm by allowing it to play against itself 8 billion times. This may be how future AIs solve real world problems: by simulating incredibly large numbers of alternative models and assigning probabilities to the outcomes.
I imagine these decision systems will gradually be designed to tackle problems with more complexity and less information, until they are ready for real-world problems.