Canadian clever clogs solve poker, sort of

9 Jan 20151 Share

Share on FacebookTweet about this on TwitterShare on LinkedInShare on Google+Pin on PinterestShare on RedditEmail this to someone

Share on FacebookTweet about this on TwitterShare on LinkedInShare on Google+Pin on PinterestShare on RedditEmail this to someone

Canadian researchers claim to have finally developed an artificial intelligence model that can’t lose at poker.

More accurately, they claim it can’t make a mistake at heads-up limit hold’em poker, just one strand of the most popular flavour of the card game in the world.

“Poker has been a challenge problem for artificial intelligence going back over 40 years, and until now, heads-up limit Texas hold’em poker was unsolved,” said University of Alberta researcher Michael Bowling. He is also lead author of the paper on the research, published in the journal Science.

“The breakthroughs behind this result are general algorithmic advances that make game-theoretic reasoning in large-scale models of any sort more tractable,” said Bowling.

“We’re not saying that it’s guaranteed to win money on every single hand,” Bowling added, “what we’re saying is that, in the long run, if you looked at all the hands that could happen and you averaged all of those, then the computer can’t be losing, at a losing rate — it has to be either breaking even or winning.”

Impressive work to get to this stage

When looking at the process that the computer programme called Cepheus went through, it’s hard not to be impressed.

It played against itself for 70 days, continuously. Storing up information on hands, bets, raises and losses, and it learned from that what to do, or more importantly, what not to do.

Through storing up a large database of scenarios, it learned what would happen if it raised (and won), in comparison to calling, lengthening the hand, and making more in the long run.

“That amount then gets stored as a regret value,” said Bowling in The Verge.

“And so it computes that regret number for every single action, of every single place that it gets to make that decision.”

Eventually, claimed Bowling, Cepheus reached the standard of “perfect play”.

Plenty of research went into polishing up this poker programme. Image via University of Alberta

Too good to be true?

I’m dubious, and let me explain why.

I’ve never beaten my uncle at Connect 4, not once. He maintains the key is the two-times multiplication tables, and ever the pro, he has refused to explain to me how he never loses. I’ve gotten the odd draw here and there, but that’s it.

I’ve played a friend at chess who has often beaten me by the third or fourth move, knowing how he will win and just biding his time.

Both these people could be destroyed by an artificial intelligence model relying on basic, but still comprehensive, equations. Every move is clear to see, nothing is hidden, everything in front of you is reality.

Poker is only partly to do with reality. Another, large integral part, is make believe. People believe one thing, even if the cards say another. Also, you rarely, if ever, get to see all the cards in play.

To infinity and beyond

Also, this model works ‘in the long run’, working out its moves in extremely unrealistic circumstances. “Imagine someone playing 200 hands of poker an hour for 12 hours a day without missing a day for 70 years,” said Bowling.

“Furthermore, imagine them employing the worst-case, maximally exploitive, opponent strategy, and never making a mistake.”

Unfortunately, ‘infinity’ simply isn’t an ingredient in poker, while playing a hand perfectly is only ‘perfect’ if you are playing someone of a similarly high standard.

That rarely exists in real life, which may render these findings, essentially, a flop. But judge for yourself.

Pocket aces playing against a computer image via Shutterstock

Gordon Hunt is a journalist at Siliconrepublic.com

editorial@siliconrepublic.com