Community Discussions
Explore the latest discussions and community conversations related to this domain.
Poker Training AI
Main Post:
Hello.
I want to create a poker training AI along the lines of PokerSnowie, but better. I don't know AI very well, but I do know poker. Can anyone recommend a forum where I might be able to find a RL expert to partner with?
Thanks,
Britt Griscom
Top Comment: This is the first time I've heard of Pokersnowie. In the AI research community the leading approach to developing strong AI is similar to the following bots: deepStack (University of Alberta Heads-Up), libratus (CMU heads-up), and Pluribus (CMU multiplayer). All of these approaches use ideas similar to reinforcement learning but really they are doing what is called self-play no-regret learning. Of course there are many more details, but essentially no-regret self-play allows you to compute a near perfect strategy in the two player case. For more than two players it has been observed to work well empirically (Pluribus) but we don't have a good theoretical reason for why this is the case. Recently there have been attempts at methods more like reinforcement learning, although i don't think they have been able to reach the performance levels of the mentioned bots (maybe because they have yet to be tested against pros or trained on the full size of no limit poker).