Champions Declared In AI Poker Tournament 54
the_newsbeagle writes "The annual computer poker competition has just wrapped up, in which artificial intelligences battled each other over the (virtual) Texas Hold 'Em table. A researcher who worked on one of the top programs, the University of Alberta's "Hyperborean" program, has blogged about this year's competition and entrants for IEEE Spectrum. His first post explains the rules of the game and why it's tougher for a computer to win at poker than at chess; his second post describes Hyperborean's strategies, and the third gives the results and takes stock of Hyperborean's performance."
Mindgames (Score:2, Insightful)
According to the human players, poker is largely about mind games.
In AI poker, the competitors should be able to send files to each other, or somehow exchange non-game information.
Re:Mindgames (Score:5, Insightful)
According to the human players, poker is largely about mind games.
Really? The pros I have heard have spoken about probabilities, trying to determine your opponents' strategies (are they betting conservatively? do they bet on weaker hands more frequently than expected?), and measuring expected returns quickly. Tells and psychology seem to be a small part of their strategy, and unsurprisingly, professional poker players defeat the AI players despite the fact that computers have no psychology to play against.
Re:So coming back to the age-old question (Score:4, Insightful)
Texas Hold'Em is a game of statistics. In any short-term run, luck might triumph over skill. But if you play long enough, there are a variety of strategies that consistently prove to be better than naive play. The simplest one to model is deciding whether to go all-in pre-flop. That's straightforward enough that papers like Universal statistical properties of poker tournaments [arxiv.org] have worked on distilling it down to a simple function.
My favorite example of odds-based play involves completing a flush. If the 3-card flop comes out, and you have 4 cards to a flush, the chance you will complete that flush is 35%. Many new players think "I have a 1/4 chance of getting a card of any one suit each time, so the odds I'll finish this flush are 50/50". That's wrong; it doesn't take into account that you already have 4 of the 13 cards in the suit. You have to play a fairly large number of hands to distinguish that the odds are really closer to 1/3 than 1/2 though. That's why someone who is betting based on an incorrect assessment of odds will bleed money over time to someone who bets appropriately, the edge of skills over luck here. It is a long-term edge though, and luck dominates the short-term game.
David Sklansky's writing is a good starting place filled with statistics based poker observations. The Theory of Poker [amazon.com] is the standard text on odds-based play. Sklanky's career training was as an actuary, which is one reason his numberic analysis of the game is so strong.