(Poker) Never piss off people at a jackpot Bad Beat table...

[sterlingice]

Originally Posted by Pumpy Tudors

Before I try to address the rest of your post, I need to know where you're coming from with the above statement. Are you suggesting that winning at poker is "significantly based on luck" or do you mean something else?

No, the hands are significantly based on luck (which is how it's different than chess). Tho, yes, there's something to be said for the entire component of poker being based on luck. The best player in the world, human or computer, can't win if it's dealt 2/7's all day long.

SI

[Pumpy Tudors]

Originally Posted by sterlingice

No, the hands are significantly based on luck (which is how it's different than chess). Tho, yes, there's something to be said for the entire component of poker being based on luck. The best player in the world, human or computer, can't win if it's dealt 2/7's all day long.

OK, I see what you mean. In that case, the outcome of every hand is going to be rooted in chance (or luck, if you wish to call it that). Still, the AI can calculate what its chances are of making a certain hand, and it could play accordingly. If the computer could calculate its chances of making a certain hand, and if it could calculate the chances of its opponent(s) making certain hands, then it could use those probabilities to determine what the proper play is (raise, call, or fold).

The point wouldn't be to have the AI win every single hand that it brings to showdown. After all, you don't "win" at poker by winning each hand. To me, the difference between poker and chess is that chess has a win condition. Poker doesn't, unless it's a tournament, but that's a different animal. As I said before, I believe that a computer could generate a long-term profit against poor poker opponents (of which there are many). I don't think it could beat better opponents, however, because a good human player would be able to adapt to changing game conditions better than the AI could.

As far as the "entire component of poker being based on luck," the best player in the world can't be dealt 2-7 infinitely, so when the player gets better hands, he has the opportunity to exploit those edges for value that far exceeds the losses from 2-7's. That's the thing about luck and poker. Because there's no point at which you've "won" or "lost", the better players will rise above the worse players. Theoretically, given an infinite number of deals, we're all going to get dealt the same hands at the same frequency. It's just a matter of how we play those hands that will determine who gets more money out of it.

Obviously, that's basic statistics, so I'm not trying to teach you anything. I get that you already know that stuff. My point is just that I believe a good human player is more capable than a bot of exploiting an edge for maximum value. In a general sense, I agree with SirFozzie that, at least against good competition, a bot would win small pots and lose big ones. I think that's because a poker bot would simply fold too much.

[larrymcg421]

Luck is a very huge factor in poker. If I took on Kasparov in Chess, we would know who the better player was in about 5 moves. If I took on Ivey or Lederer, I might get hot and take a few pots from them (even win some head to head matchups) and it would take sometime before their advantage became apparent.

I think a computer could be programmed to do very well at $10+1 and $20+1 tourneys, but at the higher levels players vary their play too much for the computer to detect a pattern. And a computer would get killed in late game bubble play.

[SportsDino]

You can not apply a 'Deep Blue' like algorithm for chess to poker. Deep Blue other than some useful heuristics to reduce the search space, is dependant on a monster handmade hardware calculator that chugs through possibly millions of board configurations for EACH MOVE. In chess the possible moves, and therefore the branching factor, is finite, although quite massive when you consider that the search space grows exponentially.

Poker has far more variables, and some dramatic differences. First, the 'board' is a fuzzy state, there is uncertainty as to the configuration of your opponents cards, and with multiple opponents it gets even crazier. Second, there are no finite moves in no-limit, you would have to limit it to limit poker to do any sort of search algorithm. And all such algorithms would suck given the uncertainty and nearly 47 different possible cards that can come on the turn.

So chess is an EASIER game to solve than poker, just if you look at the mathematics involved. And similar algorithms will not work, something new will have to be made to analyze poker.

-----

That said, I love AI and would like to get a PhD off of making a program that can play (and learn) strategic economic games like poker, so i don't think it is impossible to craft a system that can compete on a world champion level. Cash games might be impractical to crack first, tourney play allows you to make some critical assumptions to make poker into a game you can approach with game theory. But the corresponding game theory does not exist to handle poker yet in my opinion. I have made a sort of program to play poker, but it requires a human (me) to play it, as I fill in numerous parts (particularly putting opponents on a hand). Also the program is more of a suggestion to play, managing certain risk factors in an overall tourney, but by itself it still needs a coach to manage the game, and a player to make the play for each hand. My program is like a playbook in that regard.

But anyway, poker is complex, you can probably script a good tight limit machine that will avoid losing too much money and cashes in against fools when it has power hands, but against talent it could be busted out. An aggressive poker machine would be easy to beat if it was scripted (and search is impossible), as you can see if you play any poker sim out there (the aggressive ai's are pretty moronic).