![]() |
#11
|
|||
|
|||
![]()
[ QUOTE ]
I don't think computers will ever be able to dominate a table of human players at no-limit. That's a different problem. [/ QUOTE ] I think it's inevitable that they eventually will be able to beat any human at all forms of poker. In fact I'm fairly sure that eventually they will be able to beat humans at all forms of board games, card games, and even computer strategy games (which are much harder for AI due to their complexity). I don't claim to know when this will happen, but just looking back over the last 50 years it seems absolutely inevitable it will happen sometime in the future (assuming humans haven't wiped themselves out by then). The main stumbling blocks along the way are: a) Lack computing power. b) Lack of applicable algorithms. When it comes to (a) I don't think anybody can deny that computers are advancing rapidly and Moore's Law seems to be holding. I'm fairly sure that in 20 years the same will be true and we will be looking at computers with over 1,000,000 times more processing power than we have currently (possibly much more if some huge breakthrough occurs). We are just beginning to approach the amount of computing power needed to make a serious attempt at Poker AI and a lot of the work that has being done recently would not have been possible using hardware from even 10 years ago (ie: you would be looking at prohibitavely long computational times and prohibitively expensive memory resources). Even so, I don't think that even with the predicted computing power available to an academic in 20 years time poker will be anywhere near "solved". This leads me onto (b). I agree that machine learning seems to have stagnated over the last 15 years or so (with most significant breakthroughs coming from applying well known algorithms using better hardware), but I don't think this necessarily means it will stagnate like this forever. As with all past scientific breakthroughs and inventions; all it takes is one smart guy to have a "eureka" moment and what once seemed impossible becomes reality. I also think that the increased computing power of the future will mean that future machine learning researchers will be able to test out their ideas at a much faster rate than they have been able to in the past (I myself have been involved in AI projects that have taken weeks of CPU power; only to look back and realize that I could run the same in a few hours on my current desktop PC). When it comes to Poker AI specifically, I think one of the main stumbling blocks is the "get rich quick" mentality which is stifling current research. This not only applies to the wave of bot-builders trying to cash in from using their bots online, but is also evident in academia (eg: the lack of repeatability in the UofA papers and the subsequent commercialization of their work in the "Poker Academy" software). Past research on games like chess did not have this problem and I think the sharing of information means it was not as handicapped by greed as Poker AI is currently. I'm quite sure that if all the academic man-hours that were aimed at computer chess were somehow re-applied to Poker AI then we would have some very strong poker engines available today. In summary: to begin to compete with humans we have to try catch up on a billion years worth of multicellular evolution spread over the vast surface of the planet. It's not going to happen overnight, but it's almost certainly going to happen sometime in the future. Juk [img]/images/graemlins/smile.gif[/img] |
#12
|
|||
|
|||
![]()
[ QUOTE ]
1. There is also a luck factor in chess, and in almost any other sport. 2. Poker is a game of imperfect info, but that's not a problem at all for a computer. I don't see the point here. [/ QUOTE ] knock up a program then |
#13
|
|||
|
|||
![]()
[ QUOTE ]
no, chess is a game of perfect information, wheras poker is one of imperfect info. In chess, a computer need only see the pieces to be able to find the optimal move, wheras with poker this move is subjective to many variables, not all of which are known. To find this optimal poker move you need to be able to LEARN - humans can learn, and to the extent needed, computers can't (yet). [/ QUOTE ] In theory almost the same search algorithm that is used for games such as chess and backgammon can be applied to poker. Just as in these games you could in theory expand every single action and chance node until you have the entire game tree and then back up the expectations at the leaves to find the "optimal" move ("optimal" in terms of game theory). Sadly this isn't possible for all but the most trivial games, so we have to find ways round it (ie: pruning, heuristics, stochastic sampling, etc). The big difference for poker is because it is much more of an exploitative game and to exploit your opponents you need to have a model of how your opponent behaves. Creating this model is what adds the extra complexity and this is what humans have evolved to be very good at. If you were to analyse how a great poker player thinks, then I'm quite sure they prune the game tree quite early and use a lot of heuristics to help them, but the big advantage they have is their ability to model their opponents (to both induce the holdings of others and also to predict which sequence of future actions is most likely to exploit the other's weaknesses). Interestingly, and only recently, it's been found that even games like chess benefit from a certain degree of opponent modeling. This can be seen by the introduction of Endgame Tablebases which actually allows certain n-man endgames to be played out "optimally". The problem lies with draw/loss states where if both players play optimally no better than a draw/loss can be yielded, but in reality the human may have to play an exact combination of many non-intuitive moves to get there and all other combinations lead to a better outcome for the computer. Juk [img]/images/graemlins/smile.gif[/img] |
#14
|
|||
|
|||
![]()
Is there much interest here in putting together computational
tools for poker? I have started a little bit. I started putting together efficient card generators (deterministic and random) and hand evaluators (improved over those in pokersource, I believe). I am interested if anyone knows of models for poker strategies. I have a copy of Billings' thesis from UofA. would be interested if anyone knows what good strategy models are out there. |
#15
|
|||
|
|||
![]()
There is a program that plays Rhode Island Poker perfectly already. Shouldn't be impossible to make the next step.
|
#16
|
|||
|
|||
![]()
I think that the main reason that computers will not ever be able to beat humans at poker consistently is nothing to do with poker's luck factor or an ability/inability to learn. It is simply that they are too predictable. If a bot always makes the most +ev move in any given situation, it should not be too hard for the best human players to work out quite accurately what they are holding. Poker is a game of incomplete information, yes, but top players are extremely good at controlling the information that they give their opponents. It is also a game that involves a lot of deception and I do not think that a computer could possibly get the better of a player that properly puts them to the test.
|
#17
|
|||
|
|||
![]()
[ QUOTE ]
I think that the main reason that computers will not ever be able to beat humans at poker consistently is nothing to do with poker's luck factor or an ability/inability to learn. It is simply that they are too predictable. If a bot always makes the most +ev move in any given situation, it should not be too hard for the best human players to work out quite accurately what they are holding. Poker is a game of incomplete information, yes, but top players are extremely good at controlling the information that they give their opponents. It is also a game that involves a lot of deception and I do not think that a computer could possibly get the better of a player that properly puts them to the test. [/ QUOTE ] Why would a computer have to "choose the most +EV move each time"? If anything the computer's ability to tirelessly mix up it's play is one of it's current strengths over humans. Computers are vastly better at generating pseudo-random sequences: Try playing a few games of Roshambo vs a computer using an optimal strategy... Juk [img]/images/graemlins/smile.gif[/img] |
#18
|
|||
|
|||
![]()
(Please forgive the "then" typos. I can't seem to edit my poll after creation and they should read as "within the next N years". Should have proof-read it more carefully first time...)
|
#19
|
|||
|
|||
![]()
[ QUOTE ]
I think it's inevitable that they eventually will be able to beat any human at all forms of poker. In fact I'm fairly sure that eventually they will be able to beat humans at all forms of board games, card games, and even computer strategy games (which are much harder for AI due to their complexity). [/ QUOTE ] Whereas I stand by my prediction that computers will never be able to beat a table of good human players in no-limit poker. Part of this is definitional. If the computer got sensory input from the humans, say by hooking each player up to a lie detector or using more advanced physiological sensors, the situation would be different. But I would argue that's no longer poker. Also, some day someone might learn how to grow a human brain in a jar, and train it to play poker. I'm talking about something resembling today's computers, but faster and with cleverer programs, working with inputs of the cards and bets only. Playing a table is different from playing a hand heads-up. If all the other players conspire against you, you can't win however good you are. I think game theoretic play based on heads-up principles will naturally push all the other players to play to your disadvantage. I don't think anyone has any idea how to build or program a computer to interact with humans. Maybe someone will do it someday, but it will take a fundamental breakthrough. There's a lot more to poker than odds and strategy. |
#20
|
|||
|
|||
![]()
Chess: A game of perfect information. You know their moves. They know yours. You can see several moves ahead.
Poker: You can't really teach a computer effective Level 2 thinking or beyond. Sure, if you have a player who's playing a little too agressively, the computer can start to reraise and slowplay. But computers will never know if the limper in early position is trying to trap with aces, or get in cheap with 54s. They will never know if the check-raise when the third heart on the turn hits means they have it. They won't be able to put you on a range of hands. Until they can play effectively more than their own cards, they will never dominate beyond the schmucks that keep hanging themselves when computers play passive. |
![]() |
|
|