master level bot created?

[The once and future king]

[ QUOTE ]
Head. Sand.

[/ QUOTE ]

Head. Arse.

[CRF250X]

[ QUOTE ]
Head. Arse.

[/ QUOTE ]

Ah, I see you're in deep.

[Innocentius]

[ QUOTE ]
unlike other games where computers have proven to be equal or supieror(sp?) to humans, i.e. chess and backgammon, poker has a large factor of "human" element. computers will never be able to match the human element. they cant think, only process.

[/ QUOTE ]

Yes, I was actually thinking about chess, where it makes more sense to talk about playing at master level, since it is actually well defined. There are master titles, and if you can play even with titled players, you could say that you play at master level. In poker, the meaning is unclear.

However, I don't agree that it wouldn't be possible to ever create a bot that plays (online) poker as well as any human. This is exactly what was said about chess computers until the late 80s. The computers lacked the necessary "intuition" and "ability to think" to play as well as a human. Later development has shown that there are ways to compensate for this.

[TheIrishThug]

i have no doubt in my mind that, at some point, there will be a computer vs human heads up match just like the chess one. there is also a strong posibility that the computer could win. the computer will "remember" all the plays the human made. the longer they play the better the computer can map out the humans style and then choose an action. the humans best shot is to end it fast before the sample size becomes large enough.

[StellarWind]

[ QUOTE ]
unlike other games where computers have proven to be equal or supieror(sp?) to humans, i.e. chess and backgammon, poker has a large factor of "human" element. computers will never be able to match the human element. they cant think, only process.

[/ QUOTE ]
That's ironic because good poker players don't think very much. They make almost all decisions very quickly using a trained ability to recognize patterns. Neural net programs often do well at this type of application. I wonder if it would be possible to design one that read large numbers of hand histories played by an expert and learned to imitate his play.

From a game theory perspective there is such a thing as mathematically correct poker play that cannot be beaten in the long run. The mathematics is too complicated to work out exactly for complicated games like hold'em, but there is no reason that programs could not eventually be developed that would defeat any human opponent in the long run using game theory.

Of course beating expert players is not solely what it's all about. You would also want your computer poker player to analyze opponents past play, identify leaks, and modify its own play to exploit them. The computer's potential to remember and analyze thousands of hands will be a big plus here.

In summary, expert computer poker is hard and it isn't solved yet, but there is no special reason why it can't be done.

[SlantNGo]

I agree 100%. The breakthrough in poker bots will come with a breakthrough in "learning algorithms", such as neural nets as you mentioned. Until then, bots will not be able to play at the level of top human players.

[ QUOTE ]
That's ironic because good poker players don't think very much. They make almost all decisions very quickly using a trained ability to recognize patterns. Neural net programs often do well at this type of application. I wonder if it would be possible to design one that read large numbers of hand histories played by an expert and learned to imitate his play.

From a game theory perspective there is such a thing as mathematically correct poker play that cannot be beaten in the long run. The mathematics is too complicated to work out exactly for complicated games like hold'em, but there is no reason that programs could not eventually be developed that would defeat any human opponent in the long run using game theory.

Of course beating expert players is not solely what it's all about. You would also want your computer poker player to analyze opponents past play, identify leaks, and modify its own play to exploit them. The computer's potential to remember and analyze thousands of hands will be a big plus here.

In summary, expert computer poker is hard and it isn't solved yet, but there is no special reason why it can't be done.

[/ QUOTE ]

[Rudbaeck]

[ QUOTE ]
unlike other games where computers have proven to be equal or supieror(sp?) to humans, i.e. chess and backgammon, poker has a large factor of "human" element.

[/ QUOTE ]

Actually there isn't a human element in online poker. At all. All your decisions are based on statistical analysis, pattern recognition and crude profiling.

Poki combined with a PT database of a few million hands would do very well in most online games. There just isn't such a bot around though.

There hasn't been all that much research effort made on poker bots. It's not exactly fair to compare todays poker bots to todays chess bots. I guesstimate that by 1965 there had been more research done on chess bots than there has been on poker bots in 2005.

Both Doyle and Cloutier claim that their superior memory is what sets them apart as NL players. But neither hold even a candle to the memory sun of computers.

I think that in the coming decade we'll see a fair amount of bots which can both hold their own against superstars and extract near the maximum against weak opposition.

[primetime32]

Let me make this comparison between poker and chess and the use of computers.

if you take the worlds greatest chess player and have him play against an average joe chess player, he will win 10 times out of 10.

However, the greatest poker player can easily lose to an average poker player (especially online). Look at the world series of poker as an example. The best players in the world havent won in 4 years. Its random new guy after random new guy. And at the WSOP the great players have the added benefit of reading players. If you took the best 5,000 chess players and had them compete in a big tourney the best players would win each and every time.

there is clearly no comparison between chess "bots" and poker "bots."

[Innocentius]

[ QUOTE ]
Let me make this comparison between poker and chess and the use of computers.

if you take the worlds greatest chess player and have him play against an average joe chess player, he will win 10 times out of 10.

However, the greatest poker player can easily lose to an average poker player (especially online). Look at the world series of poker as an example. The best players in the world havent won in 4 years. Its random new guy after random new guy. And at the WSOP the great players have the added benefit of reading players. If you took the best 5,000 chess players and had them compete in a big tourney the best players would win each and every time.

there is clearly no comparison between chess "bots" and poker "bots."

[/ QUOTE ]

I don't think this argument is convincing. First of all, chess isn't as exact a science as you imply. Good players get defeated by worse players all the time. But I agree that the variance is much smaller than in poker.

This however, does not mean that you cannot program a bot to play as well as a human. Of course the bot could still get its virtual ass kicked in any one session, but it would win in the long run against worse opponents. By your argument, you couldn't even talk about good poker players, since they don't win every time. What I mean is that I am not convinced that you cannot program a bot to play poker as well as the best human players.

[primetime32]

[ QUOTE ]

I don't think this argument is convincing. First of all, chess isn't as exact a science as you imply. Good players get defeated by worse players all the time. But I agree that the variance is much smaller than in poker.

This however, does not mean that you cannot program a bot to play as well as a human. Of course the bot could still get its virtual ass kicked in any one session, but it would win in the long run against worse opponents. By your argument, you couldn't even talk about good poker players, since they don't win every time. What I mean is that I am not convinced that you cannot program a bot to play poker as well as the best human players.

[/ QUOTE ]

Maybe good players get beat by worse players, but great players would not lose to you or me. It just wouldnt happen. Could you imagine bobby fisher losing a chess match to me? It simply wouldnt happen in 100 games.

but if you or me played against the best poker players in the world, we would win a few times.

The chess board is very complex, but it there are limited number of moves and calculations. If you move to a certain spot, the computer can calculate your next most probable move. In poker, the game can change when the flop or turn or river hit. The computer won't know whats coming and would have a much harder time adjusting to the ever changing situations. not to mention a full table of players and having to adjust to each and every players differences. Chess is heads up, so its easier for a computer to adjust.

With poker you have players getting up and new players coming in. So not only does poker have more variables heads up, it has a greater level of difficulty when playing against a full ring.

There is no doubt that programs can be made to win at poker. I just don't think it will ever be good enough to dominate the game like deep blue was able to dominate chess.