Stephen Jordan reports at the National Monitor that four of the world's greatest poker players are going into battle against a computer program that researchers are calling Claudico in the "Brains Vs. Artificial Intelligence" competition at Rivers Casino in Pittsburgh. The pros — Doug Polk, Dong Kim, Bjorn Li and Jason Les — will receive appearance fees derived from a prize purse of $100,000 donated by Microsoft Research and by Rivers Casino. Claudico, the first machine program to play heads-up no-limit Texas Hold'em against top human players, will play nearly 20,000 hands with each human poker player over the next two weeks. "Poker is now a benchmark for artificial intelligence research, just as chess once was. It's a game of exceeding complexity that requires a machine to make decisions based on incomplete and often misleading information, thanks to bluffing, slow play and other decoys," says Tuomas Sandholm, developer of the program. "And to win, the machine has to out-smart its human opponents." In total, that will be 1,500 hands played per day until May 8, with just one day off to allow the real-life players to rest.
An earlier version of the software called Tartanian 7 [PDF] was successful in winning the heads-up, no-limit Texas Hold'em category against other computers in July, but Sandholm says that does not necessarily mean it will be able to defeat a human in the complex game. "I think it's a 50-50 proposition," says Sandholm. "My strategy will change more so than when playing against human players," says competitor Doug Polk, widely considered the world's best player of Heads-Up No-Limit Texas Hold'em, with total live tournament earnings of more than $3.6 million. "I think there will be less hand reading so to speak, and less mind games. In some ways I think it will be nice as I can focus on playing a more pure game, and not have to worry about if he thinks that I think, etc."
(Score: 0, Offtopic) by kaszz on Sunday April 26 2015, @09:54PM
Expect to have to play "poker" at every airport within a few years. If you happen to be facially botox'd then you will have an exclusive TSA anal poke.. :p Oh we of course think of every minority. They will have their own culturally adjusted TSA probing..
Then it will be the schoolchildren, banks, grocery store, post office, your office etc..
Facebook will surely not miss this chance to tag every photo with liar etc.
(Score: 3, Interesting) by captain normal on Sunday April 26 2015, @10:28PM
My guess is that at first the human players will be giving up a slight edge to the software. Then after a few hundred hands they will figure out the machines tendencies and will wind up winning.
Everyone is entitled to his own opinion, but not to his own facts"- --Daniel Patrick Moynihan--
(Score: 3, Informative) by captain normal on Sunday April 26 2015, @10:42PM
Just found buried in one of the links, this: https://www.cs.cmu.edu/brains-vs-ai [cmu.edu]
Seems this started 3 days ago and the humans started strong but the computer has come back to only $59,766 down to the apes.
Everyone is entitled to his own opinion, but not to his own facts"- --Daniel Patrick Moynihan--
(Score: 1) by dingus on Sunday April 26 2015, @10:38PM
they're going to lose at first. The machine works with pure mathematics and probability, and poker has so much more to it. They're used to playing against humans, who are much more complex than a few algorithms. They're going to lose at first, but as Polk said, once they get used to it the machine will lose more often.
(Score: 1) by Ethanol-fueled on Sunday April 26 2015, @11:34PM
Humans are going to try to start confusing the machine with illogical plays much like they already game the HFT algorithms in stock trades.
(Score: 1) by nitehawk214 on Monday April 27 2015, @06:29PM
In fact, I would go as far to say HFT and poker are exactly the same thing.
"Don't you ever miss the days when you used to be nostalgic?" -Loiosh
(Score: 2) by CirclesInSand on Monday April 27 2015, @12:00AM
The machine works with pure mathematics and probability, and poker has so much more to it.
No it doesn't, not heads up poker. The only way poker has more to it is with multiple players, because poor play is a form of collusion in poker.
But as far as 1 vs 1, poker is just pure mathematics and distributions. An arbitrarily fast with arbitrary memory computer can be programmed to play perfectly, the only challenge is playing with limited hardware. But there is no mystical side to poker, it's all math.
(Score: 0) by Anonymous Coward on Monday April 27 2015, @12:41AM
A machine that always makes perfectly logical moves in poker is predictable. Being predictable is not a desired characteristic in poker.
(Score: 2) by SubiculumHammer on Monday April 27 2015, @01:05AM
AND how hard is it exactly to add a randomiser to a computer program? Not hard.
(Score: 0) by Anonymous Coward on Monday April 27 2015, @01:27AM
That's where it gets interesting as well. A computer that acts completely randomly is not going to win either.
(Score: 2) by melikamp on Monday April 27 2015, @03:34AM
This discussion gets into some really interesting places. For one, there are 2 completely different ways to write a poker player: a non-loser and a winner. A non-loser would play a perfect break-even strategy, which would involve randomizing each and every move, so that opponents cannot obtain any useful information. I have no idea if the following is true, but it seems to me that there might be a perfect strategy where a non-loser would completely ignore other players' possible motivations, would not try to guess any hands, and just derive its moves from the amount in the pot, its own chip stack, and the odds of having the winning hand (assuming some static "non-losing" distribution of opponents' hands, the same as the one the non-loser itself plays). The existence of such a strategy seems plausible to me because what's the point of "reading" the opponents if we can assume they leak no useful information?
A winner would be a completely different kind of animal. It would play a lot like the non-loser, but it would also play like a human and try to exploit the opponents' weaknesses (if any) by "reading" their hands, based on their previous plays. A sophisticated enough winner would probably crush all human players, but lose to a non-loser, since it can only try to win by deviating from the perfect strategy.
(Score: 0) by Anonymous Coward on Monday April 27 2015, @02:50PM
A perfect strategy, playing solely based on the 'facts', makes the machine too predictable. In poker that's not going to get you far.
(Score: 0) by Anonymous Coward on Monday April 27 2015, @03:45PM
You could switch the AI each hand, randomly, and no one would know if they were playing against the 'pure logic' AI this hand, or one of the the 'read the players' AIs, I would think being able to switch without a tell would be pretty powerful.
(Score: 0) by Anonymous Coward on Monday April 27 2015, @04:50PM
That's what makes poker much more interesting than chess when it comes to AI. A randomizer could easily cause the computer to lose just as well. How do you make a computer unpredictable enough to confuse opponents while still, on average, being able to make more money than it loses. and an AI that does nothing to attempt to predict its opponents but has opponents with the intelligence to predict it to some extent doesn't sound like such a great AI.
In poker you won't win every hand. Winning and losing is partly based on the hand you're given. The computer may lose simply because it was randomly given a worse hand than you even though its hand is statistically favored. What makes for a statistically good poker player, after playing a more statistically significant number of hands, is a lot more than just playing based on a predictable statistical algorithm. The last thing you want to be is predictable but, on the other extreme, you don't want to act completely randomly either because that will also cause you to lose. There is a lot of intelligence involved as well. Money changes hands much faster when people start betting. If the computer acts randomly on a big hand it may lose a lot of money which could cost it the game. If it's programmed never to act randomly on a big hand that makes it predictable in the long run which is also bad. One must also factor in how much money you have vs how much money your opponent has. An computer with relatively much more to lose than you may attempt to act more conservatively. However turning off the randomizer during these times also makes it more predictable. Turning it on may cause it to lose very important hands. Making it random only during smaller, less significant hands, when the computer has relatively more to lose also makes it predictable. Poker is much much more than just sticking a randomizer next to a statistical algorithm.
(Score: 0) by Anonymous Coward on Monday April 27 2015, @03:16PM
"I have no idea if the following is true, but it seems to me that there might be a perfect strategy where a non-loser would completely ignore other players' possible motivations, would not try to guess any hands, and just derive its moves from the amount in the pot, its own chip stack, and the odds of having the winning hand (assuming some static "non-losing" distribution of opponents' hands, the same as the one the non-loser itself plays). The existence of such a strategy seems plausible to me because what's the point of "reading" the opponents if we can assume they leak no useful information?"
Any poker player with any experience easily has a general good intuition of the odds of winning a given hand and how they should act, given the current situation, based purely on statistics. Any professional poker player worth his or her salt knows the statistics of each hand very well (it's rather easy to compute). But always acting on such intuition will make you predictable and a predictable player does not make for a good player in poker at all. If others can predict that you will always act the same purely on facts they will easily predict your probable hand based on your actions and they will easily act accordingly.
(Score: 0) by Anonymous Coward on Monday April 27 2015, @03:29PM
In poker you don't want your opponent to know your hand. That would defeat the whole purpose of the game. If everyone knew everyone's hand the game would be very boring and there would be no point. If a machine acts purely on some statistical algorithm that machine will be predictable, people will easily be able to predict what hand it has. That makes for a bad poker player because people will simply fold when they know you have a good hand and you will fold when the odds are against you even if your opponents have worse hands and are simply bluffing (and if you fold your opponent essentially wins the round and takes the jackpot).
(Score: 0) by Anonymous Coward on Monday April 27 2015, @01:10AM
Then, it is completely logical to not be predictable, and perfect play, by definition, would factor that in.
(Score: 1) by dingus on Monday April 27 2015, @03:48AM
oh, this is just 1v1? Sorry, I'm not well versed in poker terminology.
(Score: 2) by CirclesInSand on Monday April 27 2015, @12:06AM
There is a big difference between play 4 top poker players, and playing 1 top poker player in 4 different games. Heads up, if the computer is playing anywhere near an optimum algorithm, the humans won't stand a chance.
But if a computer were to play against 4 humans, it would probably require actual AI and knowledge of human psychology to win, because 4 humans playing non perfectly are giving each other an advantage over a perfectly playing computer. It's the same reason you see amateur players going all in against pros -- because it's their only way chance of winning. Two or three amateur players can beat a pro if they just agree to all 3 going all until it becomes 1 pro vs 1 amateur with many times the bankroll. Bad play is collusion, even if it is unintentional.
And for that reason, the computer won't play 4 pros at the same game, but in different 1 v 1 games, because then "math" is guaranteed to win without a need for psychology. If the computer loses, it really is just because it was using a bad algorithm, not because asymptotically perfect algorithms don't exist.
(Score: 0) by Anonymous Coward on Monday April 27 2015, @12:53AM
Poker is not chess. It's possible to lose simply because you were dealt bad hands no matter how good you are. On the other hand a perfect machine that always acts statistically will be predictable. I will fold when it raises. I will raise when my hand is very obviously a statistically superior hand. If the machine has a statistically good hand but mine is better I could win. If the machine has a statistically bad hand but mine is worse I could bluff and still win because it folded. A predictable machine does not make for a good poker player.
(Score: 1, Insightful) by Anonymous Coward on Monday April 27 2015, @01:05AM
And an optimal algorithm is predictably optimal. That is part of what makes a computer playing poker far more interesting and difficult than the usual CS problems.
(Score: 0) by Anonymous Coward on Monday April 27 2015, @01:24AM
Exactly!
(Score: 2) by hendrikboom on Monday April 27 2015, @12:30PM
There's a very simple game that shares some of the properties of poker, when it comes to machine playing.
Two players, E and O.
They each choose whether the turn a penny heads up or tails up.
They reveal their pennies.
If they are the same, E wins. If different, O wins.
Optimal play against an optimal opponent (the kind studied in game theory) is to be random, with equal probability of heads and tails.
But playing against a human, who chooses instead of flipping a coin, the computer can do better. It has a distinct edge in that it can determine the patterns in the human's attempt to be random. Just keeping track of what the human did after the last time there was a HTHHH sequence (and other such) is enough for a significant edge.
-- hendrik
(Score: 2) by mtrycz on Monday April 27 2015, @01:30PM
Obvioulsy, knowing this, the human will just choose to flip the coin. Why shouldn't he?
In capitalist America, ads view YOU!
(Score: 2) by CirclesInSand on Monday April 27 2015, @06:32PM
I've seen this in a lot of places, so to not respond to any one particular poster, I'm just going to write this here.
If you don't know what Nash Equilibrium is, then you shouldn't be commenting on the potential effectiveness of a computer in poker. If you don't know what probability distributions are, then you shouldn't be commenting on the potential effectiveness of a computer in poker.
One vs one, or heads up, there is nothing magical about poker. It is just as solved as chess or go, the only challenge is optimizing the algorithms so that they can run on modern computers. They are intractable, not unsolved.
What makes poker interesting from a theoretical point of view is playing with multiple players. There collusion (even unintentional) becomes a really hard concept to define.