Chess vs. Poker vs. Computer

[StregaChess]

[ QUOTE ]
"This is what seperates poker from chess.. a chess computer can beat a weak player just as effeciently as a grandmaster can."

Because there is no bonus for beating somebody badly. But say there were extra points for checkmating in less than say 45 moves. If tournaments were comprised of good and bad players alike, a computer would lose to humans it could beat head up.

[/ QUOTE ]

The only problem is this is just not true. I'll have to repeat this I guess.. COMPUTERS CRUSH WEAK PLAYERS... with greater ease than human do.....

[FortunaMaximus]

Oh, Strega, you'll have to make exceptions for Fischer and Tal here, I believe, and maybe Capablanca. IIRC, Fischer destroyed his opponents, some of who were highly regarded in their own right, in a multi-board match, no?

I'd rank that as a pretty good gauge of whether Fischer could beat Deep Blue.

[StregaChess]

[ QUOTE ]
Oh, Strega, you'll have to make exceptions for Fischer and Tal here, I believe, and maybe Capablanca. IIRC, Fischer destroyed his opponents, some of who were highly regarded in their own right, in a multi-board match, no?

I'd rank that as a pretty good gauge of whether Fischer could beat Deep Blue.

[/ QUOTE ]

No not really, I played Tal in a simul and crushed him, he's only human. Kasparov played multi board matches as you call it with much higher players most of them International Master level and won. Fischer, Kasparov or Tal have never crushed their peers in a simul, it never happen. The point I was trying to make it that folks are tossing around that computers beat weak player slower than master do and that’s just false. Computers excel at tactics, and exploit any small tactical advantage over weak opponents, humans tend to play positional concepts or “flawed tactics” and in most cases a computer will crush a weak human quicker than a solid master would, that's just the way it rolls.

[FortunaMaximus]

Fair enough. But it's complete-information game, right? I'm not sure the same can be said for poker. Hell, it's theortically possible, I know I can mimic AI's in simulations. To a certain extent.

Hmph. It'd make an intriguing problem for the right programmer, but the Ungar gin example might be indicative of how truly difficult it is for a brute-force computational box to overcome the thinking process.

However, that's a rare breed of individual.

[StregaChess]

[ QUOTE ]
Fair enough. But it's complete-information game, right?

[/ QUOTE ]

Yes it is but not everything is known at any given moment. There are possible lines that go out side the scope of what can be known in the “moment.” I forget the particular ending but decades went by before a computer figured that some combination where it’s bishop vs. knight (and what other pieces I don't recall?) is a brute force win in something like 325 moves!

Since the OP put chess and poker and computers in the same sentence I think it’s worth really describing what that looks like. This is what happen more or less in chess and I’ll use poker as an example. First you are not asking if a computer can be programmed to play better poker than a human. That’s question does not exist in a vacuum because you are faced with money, big egos, large corporations, cheating and a drive to get it done without any boundary conditions. Makes no sense… OK… hang with me for a second (I’m tossing in poker for more or less what actually went down in chess)

1st the poker programming team will get a player to flip, they will hire a Chip Reese or someone as part of their team. They will study everything known about every top player, they will invite many (50 to 300?) of the top players in to the world to a million+ freeroll with the only condition that every hole card is captured on camera and you must submit to a brief interview a week after the event. The players will of course come on board, +EV, appearance fees too and what can it hurt it’s only a computer.. muahahaha..!?

Chip and team will start the frame work to build custom algorithms’ to beat key players, private man vs. computer poker matches will be held, again custom algorithms will be used and possibly adjusted as play goes on if the “team” determines anything out of sync. Structure conditions will be studied, we now know that Hellmuth plays 20% less optimally if blinds increase in 30 minutes as opposed to 45 minutes. This entire specific targeted BS goes on at the same time they are actually programming a computer to “play” the game. Possibly a fuzzy logic systems is tested and let run loose online, not as the final machine but to data mind for new information, maybe this fuzzy system learns something “most” poker experts to be true is false. All of his and the computer will have every table, every known situations, every bit of history it its system at all times. It will know to adjust the calling range of a Johnny Chan vs. any Phil, in any given situation and if they are wrong they’ll adjust again and again.... At some point real poker players won’t care because its such a “planned project” that you’ll still wonder if machines play better poker than humans in spite of the fact that the program has won the WSOP three years in a row, or has beaten a Phil heads up... it just won't matter or prove that much.....


Back to chess for a moment, Deep Blue team most likely had human intervention at key points……

Game Over (Deep Blue Film)

The move....

[FortunaMaximus]

It depends, actually. There is a theortical possibility that IBM may have been successful with some of its cybernetics research.

Whether this was an accidental emergent process or it was forced-emergent is a matter of debate. If it's a runaway multiple=emergent, then IBM wasn't the causative effect, and Deep Blue had human intervention.

So I'm inclined to think that there was cheating. Kasparov might have been right, and there is a very high probability that he is correct.

It depends. Is the mass information flow and storage capacity of public nets and information churn enough to trigger a multiple-emergent? There isn't even a theortical quantity to base a sample on. The only comparsion we have is the human brain, and that hasn't even been analyzed to the core, much less built successfully as far as successful neural net prototypes go though.

So, yes, they probably had human intervention, but this cannot be taken as a 100% probability.

A Clarke posit in 3001 states that it takes a petabyte to hold the whole of a human lifetime. I believe the appendix cited a source of study where this was postulated in science. I don't recall the study, though, or the cited author, so... I don't know.

In the wake of such an eventuality, a Turing test would be insufficent to draw conclusions.

It's an interesting situation though. Realize that we're no closer to knowing what trigged our own self-awareness, much less how to model a situation in which one might emerge.

Hmm.

[bearcats54]

[ QUOTE ]
i disagree with you, there are more ways to check mate, a person then there are grains of sand in all the beaches in the entire world, if some one could design a computer to take that into consideration surely they can design a computer to handle all the straight flush fullhouse etc. etc. possiblities ches is way more complex than poker poker is all bout mathmatical probalities the probalitiy your rockets will hold up againts 10 people in the game, the probaility that half will fold, the probaility the aces are good if the board hits 3 of the same suit or 3 cards in sequence, there is not that much to consider in poker as it is in chess. my point is if they can design a computer to beat a human in chess surely they can design one to beat a human in texas holdem.

[/ QUOTE ]

i think there is a lot more to poker you dont know about...

[David Sklansky]

My statement assumes both chess and poker computers are using the strategy that has the best chance against experts.

In the poker case that means it will try too many bluffs against a sucker.

In the chess case it means that it will assume the bad player will find the best move against it.

Stuff like that.

If you want to postulate a computer that deviates from optimum strategy when it sees bad play you run into the problem of counter strategies designed to make that occur.

A counter counter strategy would be nightmarish to program, I think. Practically, the best one could hope for would be a program that plays optimally. One that would not crush bad players to the degree an expert human would.

[bigpooch]

Well, StregaChess' post on "the game" is a good summary of
the unusual events. Sure, I could be wrong about human
intervention, but I'll let others who have looked into this
carefully (if any) speak for themselves.

About chess analysis, it's true that every bot will look at
all positions up to a certain depth. The only problem is
that the "evaluation function(s)" are based on mechanical
calculation. A true world class chess analyst would have
better judgment, i.e., his "evaluation function(s)" is
better. Also, most world class players have a very good
"opening book", especially world class correspondence
players. Opening theory isn't as cut and dried as many
chess players might think.

Also, in the scenario I described, it's not really a very
good test of chess excellence for several reasons, so
apologies for poor experimental design! [img]/images/graemlins/smile.gif[/img]

1) I assume all the participants would know something about
their opponents: their name, whether they are bot, previous
correspondence games. It's pretty clear everyone would
treat a bot as if it would play like a bot, etc., so there
would be some advantage to the human participants.

2) Well, I hate to bring this up, but the humans would have
a HUGE ADVANTAGE a la Russians versus Fischer, only now
there would be two "victims"!

3) Also, depends on where these bots originate from: if one
of them was a huge corporation, I think you might get the
picture.... (ok, it's even uglier than 2 above!).

[AlienBoy]

[ QUOTE ]
if some one could design a computer to take that into consideration surely they can design a computer to handle all the straight flush fullhouse etc. etc. possiblities ches is way more complex than poker poker is all bout mathmatical probalities the probalitiy your rockets will hold up againts 10 people in the game, the probaility that half will fold, the probaility the aces are good if the board hits 3 of the same suit or 3 cards in sequence, there is not that much to consider in poker as it is in chess. my point is if they can design a computer to beat a human in chess surely they can design one to beat a human in texas holdem.

[/ QUOTE ]



I think your statement is completely wrong for exactly the reasons I attempted to infer in my original post.

Chess can be solved by brute force, which is largely what Deep Blue did. It can only be solved by brute force however because it is a game of nearly complete information. I say "nearly complete" because you don't know what your opponent is *thinking* - though you do know the available possible moves, and you can analyze thousands of move branches.


In poker, and most especially in multi handed situations, you have a fraction of the available information that you have in chess, so brute force analysis is impossible.

And much of the available information is subject to manipulation.

E.g. Is seat 3 betting with AA or KK? Or is he representing AA or KK? Is seat 4 overcalling with a draw? Or slowplaying a set?


Added on top of the uncertainty of available information, is the randomosity of the progress of the hand, and in hold'em the ability of certain cards to improve your opponent more than it improves you, and which is even more challenging in multi handed situations.

Now while some of this can be "poker stoved", much of it relies I think on human intuition.

For myself personally, in a live game my advantage is in my reads. I feel comfortable making certain plays as +EV when I can put my opponents on a specifically narrow range of hands. Even though that play is -EV for a wider range of hands.

I do not believe a computer can make such a judgement call on a human.

If I was playing against a computer, I would play against them as if it were someone I could not read intuitively - but since I would expect the computer to be making +EV game theory plays, I'm pretty sure I could I could manipulate information enough (i.e. bet sizes etc) that the computer would not be able to read me accurately, either. In fact, I think if I simply randomized my bet sizes the computer wold be unable to play effectively as over time it would be unable to find any pattern to my betting.

Or a strategy might be to play very consistently, always showing folded cards - then after a certain amount of play, reverse play styles completely. There would certainly be enough lag while the computer reassessed my play style change to steal pots.


Of course as I write this I wonder if this is the answer to a functional computer model - Play one way, switch gears and play random, switch and play TAG, switch and fish for a few hand, switch and play LAGTAG, etc.



Regardless and back to my original post, I still believe that "brute force" cannot be used to solve a NL multihanded game.



AB