Poker has been solved!- news article (1 Viewer)

byline9

Pair
Joined
Nov 7, 2014
Messages
139
Reaction score
146
Location
Phoenix, AZ
Read this interesting news story today. I wondered about the accuracy of this, and what other players here think.

Computers Are Learning How To Treat Illnesses By Playing Poker And Atari


I did not think poker was solvable due to the human element. I had thought that backgammon had been solved, and that Deep Blue essentially solved chess for all intents and purposes, but this story disputes that. Perhaps I heard wrong in the past.
 
The story hit the mainstream news early last week, and was discussed either here or on CT. And it's only heads-up, limit hold 'em that has been solved.
 
I thought HU LHE was solved several years ago for all practical purposes.
 
Because the resulting strategy is largely game-theoretic, Cepheus is not beatable in the long run, but it's also not ideal against any given non-Cepheus opponent.

That is, against a given human opponent with particular flaw, a great poker player can potentially take a strategy that will exploit those flaws better than Cepheus' strategy. When the flawed opponent tries to react by changing their strategy, the great poker player can use that as a new flaw. Cepheus will continue to play the same unbeatable game as always.

In theory, better computer "players" could be created, in the sense of out-performing Cepheus versus a particular batch of opponents. But it is essentially impossible to out-perform Cepheus when playing against Cepheus.
 
So when will they match Cepheus against Cepheus and just let them run?

You'd think they'd have done that already, right?
 
In fact, pltrgyst, that's exactly how they made Cepheus.

They picked the simplest possible poker game:
- Holdem has more board cards / fewer player cards than others
- Heads-up has fewer situations than any higher number of players; you're always dealer, or not.
- Limit has fewer betting situation - check, bet, raise, fold. Never amounts. And there's no actual bankrolls; it's running scores, so all-in does not happen.

Then they wrote algorithms that remember each outcome for each possible situation (situation meaning each possible order of cards and betting order from start to finish), and figures the average result... it learns, through exposure, what decisions lead to the best outcomes.

And to give it that exposure - the chance to see all those situation - it played every deal against itself over and over. It played the equivalent of a billion billion hands against itself. It was never told what decisions were right - just that coming out ahead, on average, is better. It figured the rest out. It tried everything, lots of times, and saw what made for the best result in a given situation.

The total "strategy" is pre-calculated and saved on disk - i.e., how to play in every exact situation, including what percentage of the time to take each action (what percentage of the time to fold, call, and raise in a given situation). When I say "every situation," I mean, every situation - given these two hold cards in dealer position, with those four cards on the board, and this exact betting progression for preflop, and flop, and so far in the turn, Cepheus will now fold/call/raise this percentage of the time.

The total strategy, saved on disk, takes 12 terabytes of space, compressed. No pictures; all text and numbers. Crazy stuff... but remember, this was for the simplest stakes in the simplest player count in the simplest type of poker.
 

Create an account or login to comment

You must be a member in order to leave a comment

Create account

Create an account and join our community. It's easy!

Log in

Already have an account? Log in here.

Back
Top Bottom