Facebook AI Pluribus Defeats Top Poker Pros In 6-Player Texas Hold 'Em (arstechnica.com) 58
Carnegie Mellon University and Facebook AI research scientists have developed an AI dubbed Pluribus that took on 15 professional human players in six-player no-limit Texas Hold 'em and won. The researchers describe how they achieved this feat in a new paper in Science. Ars Technica reports: Playing more than 5,000 hands each time, five copies of the AI took on two top professional players: Chris "Jesus" Ferguson, six-time winner of World Series of Poker events, and Darren Elias, who currently holds the record for most World Poker Tour titles. Pluribus defeated them both. It did the same in a second experiment, in which Pluribus played five pros at a time, from a pool of 13 human players, for 10,000 hands.
Co-author Tuomas Sandholm of Carnegie Mellon University has been grappling with the unique challenges poker poses for AI for the last 16 years. No-Limit Texas Hold 'em is a so-called "imperfect information" game, since there are hidden cards (held by one's opponents in the hand) and no restrictions on the size of the bet one can make. By contrast, with chess and Go, the status of the playing board and all the pieces are known by all the players. Poker players can (and do) bluff on occasion, so it's also a game of misleading information.
Co-author Tuomas Sandholm of Carnegie Mellon University has been grappling with the unique challenges poker poses for AI for the last 16 years. No-Limit Texas Hold 'em is a so-called "imperfect information" game, since there are hidden cards (held by one's opponents in the hand) and no restrictions on the size of the bet one can make. By contrast, with chess and Go, the status of the playing board and all the pieces are known by all the players. Poker players can (and do) bluff on occasion, so it's also a game of misleading information.
Pluribus vs. Watson (Score:2)
on line poker is not the same as live tables (Score:2)
on line poker is not the same as live tables Where bluffing / tells are an bigger deal.
Re: (Score:1)
An Excellent, readable, technical paper (Score:4, Interesting)
I encourage others to actually read it.
Basically, it plays against itself many times. But it can also see what its other selves would have done if it had played differently. Then sees how the play turned out and feeds that back.
To do that it simplifies its internal game somewhat. A few bet types, a few hand strengths, so the combinatorial explosion is manageable.
There is also a small amount of search involved.
Obviously no tells etc. But professional players have those well under control anyway.
One idea was that, for example, in sizzor-paper-rock an optimal strategy is to chose randomly. No opponent can beat that. Although given a human opponent there might be a better strategy involving second guessing. But just sticking to random will never be beaten any human, although it might not win either.
Re: (Score:2)
Bots have already ruined online poker. Players use them to calculate odds and count cards. If you join an online game these days you can be fairly sure you will be playing against people using bots, and unless you do too you are going to lose.
It's basically become bot players farming noobs who haven't figured out why they are losing yet.
Re: (Score:1)
Re: (Score:2)
Really? (Score:4, Insightful)
A computer that has a infinite ability to count cards wins at cards? Hard to believe it isn't it?
Re: Really? (Score:1)
Just think, some day the average computer will be as small as the family automobile!
Re: (Score:1)
Counting cards has nothing to do with poker, so that concept is completely irrelevant here.
Re: (Score:2, Insightful)
Re: (Score:2)
Re: (Score:1)
The ultimate point is.... (Score:2)
games like poker are no more
Re: (Score:2)
I don't think so. The bots in this story were trained only against really good poker players. Less skilled players tend to be more erratic in their bluffs, and just might throw these bots for a loop.
Also, human players have an amazing ability to adapt. I'm not at all sure the bots will be able to hold their edge.
facebook poker app (Score:1)
Emotions (Score:2)
I don't know why they keep doing this (Score:2)
Re: (Score:2)
If it were so easy for a computer to win, why is this a story only now, decades after chess computers were able to beat the best humans?
Misinformation (Score:1)
Given this is a Facebook project... (Score:2)
One easy way to win games like this is to cheat - let the bot players share information about their cards through channels the human players can't access.
These researchers work for the same company that did research on their users, seeing if they could change user's moods by putting up/downbeat stories in their news feed [theatlantic.com]. They openly published that.
Given that this is a Facebook project, they need to assure us they did not cheat.
Very limited test (Score:2)
These bots were trained against just a few expert players. Less skilled players tend to bluff erratically, and may confound these bots. Also, human players can learn to adapt. AND this is just six-player Texas Hold'Em. It's a lot harder to win in a bigger tournament.