Science
Related: About this forumComputer program 'perfect at poker' (BBC)
By Victoria Gill
Science reporter, BBC News
Scientists have created a computer program they say is the perfect poker player and never makes a mistake.
The developers told Science journal they had "solved" the two-player game Fixed-limit Heads-up Texas Hold 'em.
And the algorithm had a strategy so close to optimal "it can't be beaten with statistical significance within a lifetime of human poker playing".
The poker-ace algorithm is also now available online for people to test, query and even play against.
***
This new poker-playing program has taught itself to overcome this. It has played trillions of hands of poker and been designed to learn by "regretting" and remembering every decision that does not lead to the optimum outcome.
***
more: http://www.bbc.com/news/science-environment-30718558
abstract: http://www.sciencemag.org/content/347/6218/122 (paywall for full article)
U Alberta Computer Poker Research Group: http://poker.srv.ualberta.ca/ (play against Cepheus)
By odd coincidence, I just finished reading the chapter on poker in Nate Silver's The Signal and the Noise, so this caught my eye.
adirondacker
(2,921 posts)I'll let ya know how I make out
hunter
(38,318 posts)I subscribe to the theory that we'll never identify the point where machines became intelligent or conscious; instead we'll begin to recognize our own minds as a very complex set of specialized (and somewhat redundant) subsystems linked together by a very flexible network.
jakeXT
(10,575 posts)greiner3
(5,214 posts)Games to be perfected.
Games, such as cards, are just a running series of mathematics and who (as in AI) better to master them than masters of mathematics.
lastlib
(23,250 posts)eom
Reter
(2,188 posts)If a three comes out an an ace doesn't, it loses. Would it fold pre-flop?
Jim__
(14,077 posts)I don't play Texas Hold-em, so I may be misunderstanding the rules as stated in the BBC article (cited in the OP):
- The two players are each dealt two cards only they can see
- On the basis of these alone, the first player can choose to bet a set amount against the other or "fold" awarding the game to their opponent
- The second player can then choose to match the bet, double it or fold
- If the bet is doubled, the first player then has the option to match it or fold
- Three further cards are laid face up on the table by the dealer
- The two players bet against each other for a second time based on poker hands they could potentially complete by combining their two concealed cards with the three on the table
- Another card is laid face up on the table by the dealer
- The two players bet against each other for a third time - but now the set amount is doubled
- A fifth and final card is laid face up on the table by the dealer
- The two players bet against each other for a fourth and final time - with the set amount remaining the same as in the previous round
I'm interpreting that set amount on the first bet to be determined by the fixed limit. I don't know exactly what the implications of all-in are. Based on my understanding of it, it seems that the computer can cover your bet, lose the hand, and if you keep on playing, you will probably lose in the long run. From an article in Nature:
My understanding of the rules is that the computer's opponent could only go all-in if he only had enough chips to cover the first bet limit. That's what protects the computer from losing one big game and so losing over-all to the opponent. Do you think that the opponent can go all-in and make a big bet that would allow him to win?
brett_jv
(1,245 posts)of the required bet. IOW, if you're betting on the turn (i.e. after the 4th shared card is showed on the table) and the game is $3/6 limit game, the minimum req'd bet is $6. If you only have $5 in chips on the table (your bank), you may go 'all-in' with your last $5, but that's very different from 'going all in' in a no-limit game ...
Modeling a no-limit poker match would be far, far more complicated than modeling a limit game.
Also the claim is NOT that the computer will never 'lose' a hand, simply that, over a given amount of time/hands played, it will end up ahead by following it's strategy of when to check/bet/raise, and when to fold.
IOW, what they're saying is that a person could not come up with a more optimal strategy for long-term success than the method by which the computer plays the game/makes it's decisions, mathematically-speaking.
It'd be nice if the article stated what the minimum count of hands would be in order for the algorithm to produce a net positive bank with, say, 90, 95, and 99% confidence.
My offhand guess would be those counts would be on the order of 50 hands, 500 hands, and 5000 hands ... but that's a sheer guess on my part. Those would represent what I'd call 'pretty impressive' hand counts for these levels of confidence, put it that way.
I also wonder what they set as the 'max' amount that the computer will allow itself to 'go down' before it quits, as normal human player would?
ProfessorGAC
(65,078 posts). . .at least in limit games, that reading the person is not relevant, at least not to the degree that Rounders and some poker players would have one believe.
May be so in no limit, but this would suggest otherwise in a limit game.