My wife and I watched Roger Moore in The Saint last night. The episode in question opened with Simon Templar sitting at a table playing poker with some underworld characters. He cleans out the head of these by bluffing his three eights (I think?) over the other man's three jacks. The opening camera work focused a great deal on the expressions of the players. Then it occurred to me: this is why, at the present level of technology, poker AI will never, really be feasible.
At the highest level, poker isn't really about hands. There is fairly little one can do about that and, what can be done, is very strictly governed by the laws of probability. If it were, computers would whip the finest poker players right now. The game comes down to each player's ability to read the person across from him. When they're right they either win big or avoid losing. When they're wrong, they lose big. Computer's cannot read human expressions. They can have various heuristics built in and they can compute probabilities until it would make any of our (human) heads spin, but they cannot understand people beyond the most primitive tendency tracking. So they will fail, ultimately. Moreover, they cannot offer the player the fun that by and large comes with the game, of trying to see how well he does know the person across from him.
Computer poker is about hands. Human poker is about people.