Bots - quote from a chessmaster to discuss

[AngusThermopyle]

[ QUOTE ]
but I have a hard time visualizing how a bot is going to decide to "sometimes" do this or that, or to know when it's static moves have been found out and alter them in a correct way.

[/ QUOTE ]

Huh?

"Slowplay Aces 12% of the time"
1. Get dealt Aces
2. Generate a random number X, 1-100 inclusive
3. If X <= 12, slowplay. Otherwise play normal.

Works with bluffs, etc. The idea that a computer program cannot "mix up it's play" is naive. It can be programmed to give the opponents rating of 1-10 in a dozen different categories, combine that with their positions in the hand and decide what % of the time to slowplay, etc for a given hand and a given position. And then roll the dice to see what line to take.

The problem is getting the programmers and players together so they can translate the various advanced concepts into subroutines.

And we already have the equivilent of poker "cyborgs". Players with massive PT databases. And how "flexible" can someone playing 8 tables be?

Hey that's quite an argument! Good supporting facts! Don't you feel like when you have to resort to name calling, you're saying "I have no idea what I'm talking about, so I'll just insult you"? Thats how it comes accross.

[citanul]

[ QUOTE ]
Hey that's quite an argument! Good supporting facts! Don't you feel like when you have to resort to name calling, you're saying "I have no idea what I'm talking about, so I'll just insult you"? Thats how it comes accross.

[/ QUOTE ]

my argument was that:

1) you created a strawman out of the earlier comments when you said "i don't think it would be easy to do x." no one had said it would be easy, just that it would be possible.
2) i didn't name call, i didn't say that i have no idea what i'm talking about, and i didn't insult you. i also don't really need to go in to encyclopaedic detail of how AI programming works in order to make my point in this thread.

from your earlier post, where you said that you didn't believe that a computer could effectively do modeled randomized/semi-randomized/modeled with previous actions taken into account behavior, it was quite obvious that you don't understand anything besides the most basic programming.

by basic programming i mean that your post implies that you think the way computers work is like a choose your own adventure book "if x go do y." this is not the way computers work. a simplistic reality of why not is something like the post a few up on "slowplay aces 12%."

c

[Sniper]

There is definately a significant amount of effort being put into designing Poker Bots that can play at a high level, with sophisticated opponent modeling!

[CORed]

[ QUOTE ]
The problem with the idea of a bot is that a computer functions like "If condition A is true, do [this]". This is only passable at the weakest games, but I have a hard time visualizing how a bot is going to decide to "sometimes" do this or that, or to know when it's static moves have been found out and alter them in a correct way.

[/ QUOTE ]

Not that hard, If condition A Do [this] 40% Do [that] 25% Do [The other thing] 35% Run Random number generator to pick which.

[itsmarty]

[ QUOTE ]
After decades of development, chess bots reached parity with the best human players.
If even 1% of funding devoted to chess expert systems went into poker expert system, it would be capable of beating world class poker players.

[/ QUOTE ]

People always behave as if chess and poker are the only games people want to beat with computers. Significant effort has gone into Go programs as well, with only marginal success.

Martin

[Girchuck]

The existance of these "cyborgs" is exactly why I think that a good poker bot is not far away.
The main elements are obtaining a large database, and processing data according to the set of rules based on the player numbers. A bot capable of working with PT databases will be able to even select the most profitable tables, which is a major factor in winrate.

[Rudbaeck]

Bots will most likely totally dominate humans heads-up in a not too distant future in both limit and no limit texas hold'em. Because that game is completely soluble using game theory.

Now as soon as you add a third player it's not been shown that it's soluble. For full ring the problem is even worse.

In the long run I think there will be bots that play better than most strategy forum regulars here do. If they will in fact be able to beat the best in the world is less certain.

And a bot that can both beat Doyle at a full table and at the same time extract more from the fish than he does is probably way further off than a manned mission to Mars.

The problem is that the math to solve multiplayer non-cooperative partial-information games doesn't exist other than in embryonic form today. So any implementation of said math is even further off.

[Innocentius]

[ QUOTE ]

The problem is that the math to solve multiplayer non-cooperative partial-information games doesn't exist other than in embryonic form today. So any implementation of said math is even further off.

[/ QUOTE ]

I liked your post, and agree with most of what you wrote. But the fact that the math to completely solve the games isn't there does not in itself mean that it's impossible to write a very good bot. Chess isn't solved either. The best engines are all heuristic based, and I don't see any obvious reason why this wouldn't work for poker. (Observe that I'm not saying that it will certainly work, only that I am not convinced that it cannot work.)

[crazy canuck]

I would also like to point out one other difference between chess and poker. Poker is played for money. So the methodology to develop a succesful bot is way less likely to be published. Academic finance suffered and suffers from the same problem. The best models to forecast prices and to price derivatives are not published. This is why many academics believed in the efficient market theory for decades. Now there are very few empirical researchers belive
in the EMH.

Same goes for poker bots. So if you were an academic, would you rather get the recognition from a few other AI researchers or make millions. Or even better you can publish your marginal results, get the academic recognition and still make millions.