Limit Texas Hold’em - what a project subject for Artificial Intelligence experts – a game with simple rules, where pattern recognition and exploitation are key (point for the computers) but where important information is unknown, and responding to randomness and incorporating risk are essential skills (point for the humans). I can just imagine the Will Smith movie of the Isaac Asimov book of the story of the great battle between man and machine over some virtual chips, a $50k kitty and side action to the tune of $100.
Where Deep Blue might have eventually bested Kasparov, I would have been willing to bet that Polaris, the poker playing program developed by researchers at the University of Atlanta, would be unable to beat professional poker-playing humans. That is, of course, said while basking in the rosy glow of hindsight. It wasn’t, apparently, such an easy ride. The set-up was such that the two human players, Phil ‘Unabomber’ Laak and Ali Eslami played a series of games of 500 hands simultaneously, during which the cards being dealt to one human were dealt to the computer playing the other one. Putting aside the differences in the human players, this was meant to remove the vagaries of the deal, making it as fair a contest as possible.
Over the course of four rounds, the humans tied, then lost to, and finally overcame (twice) the Polaris program in a victory which left Laak and Eslami tested but vindicated in their belief that poker’s surprising complexities are beyond even a sophisticated piece of software like Polaris. This is of course when faced with players who both understand the mathematics of the game, and have that instinctive ‘feel’ for it which both makes them unpredictable and able to put opponents on hand ranges. Of course, the elusive ‘tells’ so widely assumed to be such a key part of the game were completely irrelevant when playing a computer. But they are increasingly irrelevant for many modern players, who learn the game, hone their skills, and occasionally make a decent living without ever leaving their house.
Internet poker bots probably worry online players unnecessarily; imagine the extra set of problems which playing the no-limit version of the game presents. There are quite a lot of smart people, like the good folks in Atlanta, working on the subject of artificial poker intelligence, for one reason or another, but it is clearly a very challenging task and one which highlights the peculiar mix of skill and chance in the game itself and the hard to quantify set of skills which make the good player of it.