Computer beats world chess champion, moving on to poker and go
By Cyrus Farivar 
posted Dec 8th 2006 1:10AM
posted Dec 8th 2006 1:10AM
Well, it appears that our days as a species of lording supreme in the world of chess have pretty definitively come to an end. A six-game chess match between Vladimir Kramnik (pictured), the reigning world chess champion, and the computer Deep Fritz, has just concluded. Kramnik lost, 4-2 to the multi-processor version of Chessbase's commercial software in Bonn, Germany. (To Kramnik's credit, in 2002, he'd held Deep Fritz to a draw.) However, this match may end interest in further advancing the field of chess-playing computers, according to Monty Newborn, a professor of computer science at McGill University. Newborn, one of the people who organized the match, told The New York Times: "I don't know what one could get out of it at this point. The science is done." But don't think the story ends there, as Newborn added: "If you are interested in programming computers so that they compete in games, the two interesting ones are poker and go. That is where the action is." So watch out, World Series of Poker card sharks, there's about to be a digital throwdown comin' your way.
i somehow stumbled upon this old portion of the internet and felt compelled to comment upon it as there are so many simply incorrect statements in it. i really don't have a ton of credentials other than understanding the some low-level game theory of chess, the depth of math of poker, their differences, and a background in information systems, economics, and statistics. oh hai.
solomonrex - your statement may be true for limit, maybe, but not for no limit hold'em (NLHE), which even thought rules are very similar is a completely different animal. any decent (i'd say about a year's worth of study / experience) poker player could pick apart today's bot fairly easily for 100+ big blind stacks (this is referring to cash game). also, let's be clear that NLHE tournaments and NLHE cash games are two very different animals. to determine who is best in NLHE cash games, long term expectation must be assessed (overall equity in when making / calling bets). this takes LOT of hands (i've seen massively winning players go on break even stretches of 25k+ hands or so, but usually 25k will give us a good enough confidence interval to assess if someone is actually winning (we call this deviations in expectation Variance). but when you say: "I don't think human players win poker consistently, so I don't see how a computer will. How many "stars" dropped out in the first day of WSOP this year?" you demonstrate a clear lack of understanding on our topic. "stars" are stars because they have great long term expectation. because poker is a game of skill WITH ELEMENTS OF LUCK, anyone can get bad beat and bounced from a tournament even though they made the best move with respect to your opponent's specific hand (e.g. i shove AA preflop and get called my KK, he hits a king and i lose). also, because deception is an important strategy in poker, one must plan certain hands similarly to mask their true holding. one way we do this is to bet when we have nothing relative to the board to make our opponent fold (you know this as bluffing). we'll do this with several hands and situations so we have a Range of hands which are in our actual hand (in the eyes of our opponent). but when we have KK and we bet into a K8922 board after we acted weak and our opponent shoves over us for 3 times the pot (this is how much money we have left), we call, and he shows us 22, this doesnt mean we're bad players. we ran into the top of his range (with the 2nd best hand!), which is another form of getting unlucky in poker. these two aspects lead to variance. there are more. so anyway, your "stars" losing in the first round means NOTHING. nobody will EVER have enough of an edge in a standard tournament environment to win it 100% of the time.
back on topic though: comparing chess to poker is basically pointless. chess: complete information. there's no fog of war on the board. you know all of your opponents pieces, where they are, what they do, and vice versa. this is why players need to think ahead and along the permutations of such actions and opponents actions -- the strategy level has just gotten to that point where the best of the best need to be able to do that.
poker, however is a game of incomplete information. you don't know what your opponent is holding. you don't know what cards are going to come out. you bet when you think you're ahead. you check and fold when you think you're not. for the most part. there's always balancing of acting weak when you're strong and strong when you're weak, but be careful, because if you do this too much you'll lose long term. it's easier to compare poker to the stock market than it is with chess. making good decisions long term will yield good results. if there are certain market indicators which are all pointing in one direction, you invest, and then it's force majeure in your face, that doesn't mean you're not good at investing or are stupid. you got bad beat, essentially. such things will normalize after time. we also call this the Law of Large Numbers.
duerra - you are so off the mark it's not even funny. have you even ever played poker? it's exponentially more complicated than chess given the differences in game dynamics.
cry havoc - "Until a massive revolution occurs, a computer will simply never be able to process the depth of human emotion and reactions that are necessary to play top level poker". you're overestimating the importance of physical reads / tells. now, i know the media loves to play that up like that's the core of the game, but poker is a game of MATH. yes, if you put the same person against himself and one has a pro's understanding of tells and the other does not (read: same fundamentals / skills for other areas of the game), the guy with the reads will win. however, all he is doing is incorporating another dimension of information into a mathematical model which he's using to decide his most optimal action relative to his opponent.
osu-no.1 - you get +1 Internets for knowing what you're taking about. succinct (not like me) and correct.
poker's complex because there is no Perfect way to play. it's an exploitative game. which means sometimes that weird actions must be taken to exploit / manipulate an opponent. this means, in a sense, learning. computers can remember everything, but to be able to use such information predicatively is more difficult. i have no doubt that eventually computers will best humans in NLHE eventually. it'll get to a point where their fundamentals are so close to a pro's that the only edge a pro would have are reads (useless on a computer!). and once a computer WOULD be able to get reads on a human (i'd guess we're pretty far off for this, but who knows), then the human is toast.
for more on this topic, read No Limit Hold'em by David Sklansky and the Mathematics of Poker by Bill Chen.
I say I could kick the computers power supply at a game of strip candyland. I take off clothes I survive, it takes off parts it dies. I win...Yay!.
That was a really nice article you posted. I like reading your posts , Thanks .
========
Beat the Recession with PartyPoker.com . bonus – up to £250!
Copy/Paste Bonus Code WINPOKER to benefit from this offer.
http://www.partypoker.com/td/new_table/03983_UK-gbp_PP_NewTable_LP.htm?wm=3179476 ; Submit Bonus Code when Requested Good Luck!
I agree, Poker is nice, pretty to play but much more difficult to play and to win.!
Nice blog, Thanks to post!
===============
Beat the Recession with PartyBingo.com, 100% sign up bonus – up to £100!
Bonus Code – WINBINGO ; http://www.partybingo.com/index.htm?wm=3179521
Submit Bonus Code WINBINGO when Requested ; Good Luck!
come on, no one has done the "and I for one..." joke yet?
You beat me to it! The very first thing I thought was, "How long until someone drops the 'I, for one," joke....
I, for one, welcome our new chess-playing computer overlords.
... it had to be done.
I am a chessplayer and feel like stating what any decent chessplayer sees as obvious; this match could easily have gone the other way.
In a clearly favourable position with a possible win for humankind, Kramnik made a horrible blunder allowing mate in 1. Instead of a win he got a terrible loss. This kind of thing does not normally occur at this level and is an anomaly.
In the last game Kramnik was forced to play into the computer's 'hand' in order to generate winning-chances and draw the match. He played an opening he usually does not. He would not have had big problems if all he aimed at would be a draw.
So in sum; Kramnik/humankind played better chess. Had the terrible mistake in game 2 been avoided I dare say we'd seen an entirely different result.
ChessJoe: Are you saying human error is an anomaly? Isn't the winner of most games the one who makes the fewest mistakes?
I thought you people had /.
You know, considering that this guy is so good at chess, his life probably revolves around chess. I bet the poor guy will go and kill himself now.
I wonder if a Go computer will be able to play that coveted (and generally perceived to be inexistant) "Move of God"
Go Go! (get it?)
the man will beat the machine (but remember if the robot was programed to win then why would he allow himself to loose) arrg if then statements
i think, that this isn't any real comparison. The computers need massive processing power and need the concrete tactics(like in a war having a good aim), while they don't understand anything of the real strategy.
So maybe the computer won, but there is still a lot to gain for a computer.
On the previous match, although it ended as a draw. Kramnik was clearly better, he was up by 3-1 the computer only managed to draw because kramnik decided he wanted to win on the computer terms in tactics that is.
I doubt that they will ever be able to get a computer to win at poker consistently. There are so many variables in poker that a computer won't be able to take into account. Physical tells, player evaluation, betting trends, and many other elements would be difficult if not impossible for a computer to evaluate or predict reliably. Plus you aren't playing against one person, you normally have between 8 to 9 opponents. Your strategy has to change constantly during the game, I don't think a computer has what it takes to be a winning player.
bjkalski, you've gotta be kidding. Stats can be tracked as far as how a human player plays, and the computer only has to consider one hand at a time. Plus with multiple decks and everything, the computer can calculate the odds, keep track of the known cards used, and all that other jazz. Chess is thousands of times more complicated and complex than friggin' poker. The computer has to consider some of the brightest minds in the game thinking 10+ moves ahead, and all the possibilities. There's absolutely *no way* in hell that a computer with decent programmers behind it would have a more difficult time winning at poker than at chess.
It doesn't sound like you are a poker player. I'm not a chess expert by any stretch. What I meant is that poker is a different kind of game than chess. In chess you can calculate many moves ahead and you can make logical deductions about what a player's next move is probably going to be. In poker logic doesn't often apply. Players use deception in order to hide what they are doing. As a simple example a computer might assume a raise from a player preflop might be a big hand like AA or KK but sometimes a player won't raise to keep his hand hidden or he might raise with a weak hand. I think it would be difficult for a computer to interpret things like this. In the end I think a human would be able to adapt to changes in the game much better than a computer could.
I'm guessing that duerra has never played poker as he talks about multiple decks and keeping track of dead cards, both of which would matter if you are playing Blackjack, but anyone that has ever played poker can tell you is that you use one deck, and you have no idea what cards are dead at any point (assuming we are playing Hold Em and not a draw game).
That said, I have no doubt you could program something to be a winning limit poker player, but making them a winning no-limit player would be much harder as you have to determine if someone is just pushing All-In with nothing to force you out, or if you hand is good against them, even though it might be a very weak hand statistically. In Limit, it's much easier to just make the call of the bet.
I am a Chess expert,and a professional poker player.Computers will have a lot harder time mastering Poker.The case is simple chess is pure skill,and poker is not.A computer can calculate brute force in chess to find the right answer,it cannot do this in poker.
So, duerra, do you suggest that computers are used as psychological counselors as well? By your estimation, they should be nearly perfect at clinical psychology. They could measure the pulse rate, inflection, and self-esteem levels of a patient on a huge spectrum of graphs and charts and then spit out the solution.
Except for the simple fact that people are not machines. Chess is a game that can be reduced to a science. The moves are always the same. A bishop cannot jump off the board and start pacing back and forth, so there is no need to measure it. The emotional state of a chess player cannot be counted on to influence the board, due to the mathematical calculations involved in each move.
Poker may be less complex than chess from a statistical perspective, however it's focus is much more rooted in psychology and the ability to read another person. Until a massive revolution occurs, a computer will simply never be able to process the depth of human emotion and reactions that are necessary to play top level poker. Especially in a heads-up game, the computer wouldn't stand a chance. It would calculate the odds of winning and losing and fold it away when the odds were less than 50/50. You'll note that in Star Trek: The Next Generation, Data (far more advanced than any computer today) experienced the same thing, he was unable to calculate the pot odds and cohesively integrate them with human emotions. This is not to suggest that how Data fared against Riker is an accurate representation of a computer's ability to play poker, however it should be noted that poker is a game of emotion and intuition, not science, mathematics, and tactical advantage on a set plane of coordinates.
Cry Havoc I agree. I am an avid poker player. Psychology and the human element are a huge part of the game. I've played live against people that play primarily online. Those players once they get onto a live table have a much harder time adapting. I think a computer might be able to beat some novice players by playing a tight game but it wouldn't be able to touch the pros.
duerra, are you on crack?
If poker was so easy to beat with computers, it would be done by now, because you're forgetting one simple thing: you can make money playing poker. Thousands, even millions playing online. You can't do that with chess.
If poker was so easy to beat with computers, there would be a literal stampede programmers turning into overnight millionaires by running poker bots.
Do you honestly think that software engineers and programmers chose to work on chess instead of poker out of some sense of honor or nobility or something?
Programmer: "You know, I could make millions with my poker bot while I sleep, but I'm just not feeling it today. Instead, I'll spend several years trying to beat some chess geek. That will impress those Engadget guys for sure."
Poker is much more difficult to predict than chess. In chess and checkers your opponent can move in finite ways, making it possible to deterministically predict their moves, and plan accordingly.
In poker, there is a much higher level of uncertainty, especially in hold em. You start out seeing a very small percentage of all the possible cards in the deck. Players can bluff or play deceptively to throw off your strategy. Poker lends itself towards probabilisitic analysis and computing techniques, and that's why the comment was made in the original article about how it's the new field. Look into the world robot poker championship - there is absolutely big money on the line and think about the money that you could make putting bots into sites like PartyPoker where lots of bad players go to waste money.
http://www.informationweek.com/showArticle.jhtml?articleID=165701734
Also they can't measure whether a computer is successful playing poker in one game or tournament. The measure of a good player is how they play over time. A top player can get a run of bad cards or bad luck and lose a tournament but say over a year they will win much more than a novice when you take luck out of the equation. I hope if they test a computer system they take that into account and test it over many tournaments but I can see it now a computer getting "lucky" against a pro in one tournament and the designers bragging about it.
So let me get this straight - the computer LOSES two games and "...The science is done."?!??!!? How about we say it's "done" when the computer wins ALL the games it plays? Since when is a 66% winning percentage considered unbeatable in any sport/game?
Cry Havoc Wrote:
So, duerra, do you suggest that computers are used as psychological counselors as well? By your estimation, they should be nearly perfect at clinical psychology. They could measure the pulse rate, inflection, and self-esteem levels of a patient on a huge spectrum of graphs and charts and then spit out the solution.
--
Computers are quite useful in biofeedback applications which work basically as you described. In many cases this type of work is done without the aid of a technician or counselor. It's not going to "print out a solution" per se, but it can be a useful tool towards a therapeutic goal.
With regards to computers playing poker, I'm not so sure that the ability to read and deduce human emotion is as necessary to success as many poker players would like to believe. Short of finding that a player has a tell, a players emotional responses will naturally be driven largely by the game mechanics, their current standing, their anticipation of the other players' intents, and their personal risk. By default, it is impossible to emotionally affect a computer player or draw emotional inferences from a computer player which is often what gives a good player a significant advantage. It is a hard problem to be sure, which is why it's mentioned as a frontier field when compared to chess-playing software.
John Laur wrote:
"By default, it is impossible to emotionally affect a computer player or draw emotional inferences from a computer player which is often what gives a good player a significant advantage."
You missed the point. The point is not that a human player will be unable to "read" a computer as they do another human, the point is that a computer will not be able to read a human.
In poker, a human may do things that are statistically stupid. In other words, a bluff.
In addition, unless the game is heads-up, the computer will be forced to contend with 3-8 people who could all potentially bluff at a given pot.
Calculating odds is just one small part of poker. Other factors include what the person did previously, what position they're in relative to the dealer, how much money they have left, etc.
Did Cry Havoc ACTUALLY cite Star Trek...?
What is this world coming to...
Multi-CPU vs. Single-Human ???
I'll be impressed when a single-core computer can beat a single human. Or for that matter, when a 500 core computer can win against the top 500 chess players in seperate simultaneous games.
for a computer to beat a go player, i say it wouldn't happen within the next fifty years. there are more possibilities in go than there are atoms in the universe. unless quantum computing is able to increase the computational power 100 folds, even the best go program cannot beat an average player
John Laur: As a student of psychology (graduating in 8 days), I am aware that computers are a useful tool in my field. However, I stress that they are a tool, and little more. If left to their own devices, I am confident that even the most advanced computer in the world would fail miserably at attempting to lend comfort to someone with depression. It is so far beyond their realm of capability at this point that it is barely worth discussing except as a distant potentality.
DS: Yes, I did, but only as an illustration, not as a concrete example. Data played perfect pot odds every time, and was rarely (if ever) successful. There are many players around in poker that often stick to pot odds throughout the course of a game. This is usually to no avail, because if you're holding a hand with a 48% chance of winning in heads-up, you have to fold. To say nothing of the complications of introducing 6 other players into the pot.
Computer programs ALREADY make money at poker, harvesting online poker games from the gullible. See, there's a difference between beating a grandmaster and a random person at chess, and there's a difference between making money off the gullible online and winning the WSOP.
Having said that, I don't think human players win poker consistently, so I don't see how a computer will. How many "stars" dropped out in the first day of WSOP this year?
I want to see a computer beat me at Risk!