I’m not familiar enough with Poker to say whether any of the differences between Texas Hold’em, Omaha Hold’em and Seven Card Stud should make the latter two difficult if the first is now feasible.
I’ve played all of these and my sense is that Seven Card Stud would be relatively easy for computers to learn because it has fixed bet sizings just like Limit Holdem, which was solved long before No Limit. Some of the cards are exposed in Stud which creates a new dynamic, but I don’t think that should be difficult for computers to reason about that.
Omaha seems like it would be about as difficult as Texas holdem. It has the same sequence of actions and the same concepts. The bet sizings are more restricted (the maximum is determined by the size of the pot instead of no limit), but there are more cards.
As far as I’m aware, none of the top poker bots so far were built in a way that they could learn other variants of poker without requiring a lot of fine-tuning from humans. It’s interesting to think about whether building a generalized poker bot would be easier or harder than building the generalized Atari bot. I’m not sure I know enough about Atari games to have good intuitions about that. But my guess is that if it works for Atari games, it should also work for poker. The existing poker bots already rely on self-play to become good.
I believe that the poker bots met the mark for two player games and that Omaha/7-Stud are both not much of an issue and wouldn’t actually be required in any way, but actually winning the WSOP requires mostly winning at 9-10 person tables. I do realize that there are claims they’ve been able to handle that, but doing it in person is… trickier. Probably need another level of improvement before they have a reasonable shot at winning.
(Note that WSOP is a good structure but even so is still pretty random, so e.g. I would win it some % of the time if I entered, whereas if I played the type of match they used to test the poker bots, my chances would be about epsilon.)
Another thing is that the bots never make exploits. So when there’s a bad player at the table playing 95% of their hands, the bot would never try to capitalize on that, whereas any human professional player would be able to make extra money off the bad player. Therefore, the bot’s advantages over human professionals are highest if the competition is especially though.
I’ve played all of these and my sense is that Seven Card Stud would be relatively easy for computers to learn because it has fixed bet sizings just like Limit Holdem, which was solved long before No Limit. Some of the cards are exposed in Stud which creates a new dynamic, but I don’t think that should be difficult for computers to reason about that.
Omaha seems like it would be about as difficult as Texas holdem. It has the same sequence of actions and the same concepts. The bet sizings are more restricted (the maximum is determined by the size of the pot instead of no limit), but there are more cards.
As far as I’m aware, none of the top poker bots so far were built in a way that they could learn other variants of poker without requiring a lot of fine-tuning from humans. It’s interesting to think about whether building a generalized poker bot would be easier or harder than building the generalized Atari bot. I’m not sure I know enough about Atari games to have good intuitions about that. But my guess is that if it works for Atari games, it should also work for poker. The existing poker bots already rely on self-play to become good.
I believe that the poker bots met the mark for two player games and that Omaha/7-Stud are both not much of an issue and wouldn’t actually be required in any way, but actually winning the WSOP requires mostly winning at 9-10 person tables. I do realize that there are claims they’ve been able to handle that, but doing it in person is… trickier. Probably need another level of improvement before they have a reasonable shot at winning.
(Note that WSOP is a good structure but even so is still pretty random, so e.g. I would win it some % of the time if I entered, whereas if I played the type of match they used to test the poker bots, my chances would be about epsilon.)
Another thing is that the bots never make exploits. So when there’s a bad player at the table playing 95% of their hands, the bot would never try to capitalize on that, whereas any human professional player would be able to make extra money off the bad player. Therefore, the bot’s advantages over human professionals are highest if the competition is especially though.