Slashdot Mirror
In New AI Benchmark, Computer Takes On Four Top Professional Poker Players
HughPickens.com writes: 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. 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, 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."
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, 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."
But can it make me a sandwich?
I think it's important to note that while we're good at making AI good at one thing, we're still a very long way from making it good at general skills.
not have to worry about if he thinks that I think, etc.
I think he's wrong on this. A computer would still need to consider what his opponent thinks he holds and raise accordingly.
Will the computer be prohibited from counting cards? Humans may bluff, but they cannot fake statistics.
If you are not counting cards, you are playing poker wrong. This is not blackjack with multiple decks.
Poker games take time (hours), people grow tired, computers don't.
People struggle at memorizing chances, taking shortcuts, computers have exact picture talking into account every single bit.
All one needs is behavior that is random enough, for human players not to guess if computer is bluffing.
Then, of course, there is luck factor, so results will fluctuate quite a bit.
In Texas Hold 'em, you see your own two cards and the up-to-5 community cards each hand, and the deck is shuffled between hands. Everyone knows what cards have been seen and what have not all the time without any card-counting skill.
Counting cards will not help
I think there will be less hand reading so to speak, and less mind games.
Isn't that what makes the game so interesting? Any good card game involves messing with your opponent's head.
It's poker that's dumb enough to be on TV. If they want a real challenge they should play seven stud.
In fact poker has rules (antes, blinds) where the whole point is to make not playing a losing move.
In a global thermonuclear war, it would be like making unused bombs self detonate in one's country after some time.
Your replies aren't quite making this clear for some reason, but counting cards is a strategy for blackjack, not poker. In poker, the deck is shuffled after each hand, so there is nothing to count. Of course, using whatever information is available to estimate the value of the hands' of your opponents is part of the basic strategy of poker that every sane player, including this computer, uses.
It's not about lying or decoying. It's about making the best decisions. You want AI that's capable of making good decisions even when the information is incomplete.
What, are you kidding me? Bluff has him ranked as number 36. That's pretty good, but a far cry from "widely considered the world's best."
I was also confused how this could work in holdem. After some googling it appears that "counting" in poker only refers to certain stud games where hands ahead of you are exposed.
In holdem the only (?) additional info you could have over a competitor is if someone in early position folded and revealed their cards (obviously uncommon). That's info that the small blind wouldn't have had, and could be very useful if you're in late position, especially on the button. But imo that's not really "counting", just gaining info to craft your betting strategy.
Decent answer here:
https://answers.yahoo.com/ques...
No, it doesnt depend. All the cards that you have seen are visible to you for the entire hand. Card counting is about remembering statistics about cards that you have seen but are no longer visible.
The guy that you linked to thinks that knowing how many outs you have is "card counting" -- no. you also apparently think so, which means that you cannot possibly have anything to add on this subject (and your ignorance on this subject is not a secret to you, so why are you pretending?)
"His name was James Damore."
Unfortunately the AI has a tell, his hard-disk lights flash when he's bluffing.
That site doesn't understand what card counting is. When you count cards, it changes how much you bet on a hand you haven't seen yet. You increase your bet (or enter the game altogether) only when the player odds are higher than normal.
Counting "outs", the number of cards in the desk that will improve your hand, is not what's called card counting in casino games.
The concern isn't about making good decisions, but about the behavior and use of the system's capabilities and emergent properties.
much of left-wing thought is a kind of playing with fire by people who don't even know that fire is hot - George Orwell
You want AI that's capable of making good decisions even when the information is incomplete.
"His name was James Damore."
Amusingly, Bluff magazine has a ranking of the best HUNL players where he's number one. But he's written the list himself :)
He really is among the best heads up no limit players though.
With a computer, its more than "counting cards" - a computer can remember every single card played and compute probabilities far more precisely than a human ever could. So the question is - how much real information can people *really* get from other players betting behavior? Is it really enough to offset the advantage from simply being able to fully compute the probabilities? Furthermore, if the computer model includes factoring in information revealed by bidding and uses that to obfuscate its betting behavior I would be surprised if this was really a good exercise in AI as much as it is simply creating a good model that can be computed more accurately than a human can do in their head.
Have gnu, will travel.
Maybe you should work out that there not all card games are blackjack. There are no revealed cards that are then hidden before a reshuffle in a texas hold em game, so there is nothing to count. Seven card stud, as one example, would be a poker game in which there are such cards - but such a small number that all the player's (well or the ones who can actually play) remember them.
TFS refers to TFA which refers to another TFA, and all of them are pathetically written. Here's a link at CMU discussing the competition. This is the second link in TFS, but it's not clear that all of the other links in the first paragraph are just trash.
In any case, a couple of points and/or musings:
As a final note: may I please encourage submitters and/or our illustrious editors to not fluff up submissions with links to crappy articles that miss most of the important points? Just the source link would have been enough - it's a good article with real information written in actual English.
Enjoy life! This is not a dress rehearsal.
I would guess that the game will just end up being more random. Humans are bad at being random number generators, so that opens the window for psychological strategies. A computer can be a pretty near perfect random number generator, and therefore is immune to the psychological aspect of the game. It could try to exploit this weakness in the human players through some kind of psychological heuristic algorithm, but I think that also opens the door to it being tricked by those same human players especially if they successfully guess what strategy the AI is using. It may just be better to go by the numbers.
Can I play you for money. Whatever stakes you want (please be a billionaire, please be a billionaire)
This would be very profitable. If I were an evil asshole with no ethics I would love to buy or rent such software out and put in a fake virtual player for each session. No way I would ever lose money unless through very odd anomalies. Oh hell who am I kidding this is how half of Las Vegas works where they hire statisticians to favor the house as much as possible.
You are dumb to ever think you can win and are just smarter than the other guys
http://saveie6.com/
Why can't a computer program do this? We have these inventions called 'cameras' and 'microphones' to supply the data. We have facial recognition software today. We have speech recognition software today. Emotion recognition software will exist someday.
RETURN without GOSUB in line 1050
I'm not scared. I think we are so far from understanding 'real intelligence' that the possibility of developing AI with significant emergent properties in my lifetime is practically nil.
RETURN without GOSUB in line 1050
The computer can count cards perfectly and brute force calculate the odds of each possible hand. The computer has no "tell"; but on the other hand, it probably can't read any human tells either. Over enough hands, the computer is always going to come out ahead, just by better calculation of probabilities. Artificial Intelligence isn't really required to give it an advantage, and other than being able to read the faces and tells of opponents, I'm not sure AI is even useful for poker. Of course, the software is probably also trying to use past bluffing history to predict when opponents are bluffing, giving this somewhat of the flavor of the Rock, Paper, Scissors programming competition (my quick and dirty algorithm sucked at that).
I've abandoned my search for truth; now I'm just looking for some useful delusions.