4
$\begingroup$

Cepheus is an artificial intelligence designed to play Texas Hold'em. By playing against itself and learning where it could have done better, it became very good at the game. Slate Star Codex comments:

I was originally confused why they published this result instead of heading to online casinos and becoming rich enough to buy small countries, but it seems that it’s a very simplified version of the game with only two players. More interesting, the strategy was reinforcement learning – the computer started with minimal domain knowledge, then played poker against itself a zillion times until it learned everything it needed to know.

Apparently Cepheus currently just plays against one person. Seeing as it managed to develop amazing strategy for this "very simplified" environment, what's stopping it from working on real/full poker games?

$\endgroup$
  • $\begingroup$ Technically, I have no idea. However, it is against the user terms of agreement to use any software to help you play the game. $\endgroup$ – JahKnows Jul 24 '17 at 15:54
  • $\begingroup$ @BenN , cool question though it needs some editing $\endgroup$ – quintumnia Jul 25 '17 at 15:15
  • $\begingroup$ @quintumnia I'd be happy to revise and improve my question. What would you suggest I change? $\endgroup$ – Ben N Jul 25 '17 at 18:52
3
$\begingroup$

The reason why Cepheus can't generalize has to do with the number of decision points.

The same authors recently let loose Deep Stack (DeepStack: Expert-Level Artificial Intelligence in Heads-Up No-Limit (HUNL) Poker) which is freaking many professional poker players out.

In the DeepStack arxiv paper, they say

AI techniques (Cepheus) have previously shown success in the simpler game of heads-up limit Texas hold’em, where all bets are of a fixed size resulting in just under 10^14 decision points.

...

The imperfect information game HUNL is comparable in size to go, with the number of decision points exceeding 10^160

...

Imperfect information games require more complex reasoning than similarly sized perfect information games. The correct decision at a particular moment depends upon the probability distribution over private information that the opponent holds, which is revealed through their past actions.

Using the same strategy for HUNL as Cepheus did is out of the question. Rather, taking an educated guess or using intuition based on the previous play (referred to as Continual Re-solving in the paper) is a method which can better handle this gargantuan game. For more information check out the DeepStack website.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.