Chess vs. Poker vs. Computer

[FortunaMaximus]

What about stochastic modeling? You could certainly be able to create that heuristically in a computer.

Although it'd be pretty much like trying to guess climate patterns. Eh. It'd spew twenties like an ATM gone wrong.

<shrugs> Certainly makes for an interesting problem.

[AlienBoy]

[ QUOTE ]
What about stochastic modeling? You could certainly be able to create that heuristically in a computer.

Although it'd be pretty much like trying to guess climate patterns. Eh. It'd spew twenties like an ATM gone wrong.

<shrugs> Certainly makes for an interesting problem.

[/ QUOTE ]


When I was a kid the best we had was a 3 day forcast - today weather models give us a reasonably accurate 10 day forcast.

But there's no "3rd level" in weather!

In poker it's what I've got, what I think the computer's got, what I think the computer thinks I've got, what I think the computer thinks I THINK the computer's got...



AB

[FortunaMaximus]

Indeed. Well, I'd rather play a table of female opponents. But, you know, that's merely preference bias. [img]/images/graemlins/tongue.gif[/img]

Now, females, they know how to count zero interrupt [img]/images/graemlins/cool.gif[/img]

[David Steele]

"However, it is a game of complete information, and thus it can be played by a computer using brute force, which is essentially what DeepBlue did in winning the match."

Usually in AI/CompSci, when some progress is made, a lot of people then denigrate the method as "nothing more than ....." Todays modern chess programs use very sophisticated ideas that took years to develop. Before computer chess was very strong, many said the same things "never never never..." See Hubert Dreyfus for ex.


But what about poker? Can a computer compete effectively against a human in a game of deception and incomplete
information?

Of course computer poker will not just use game trees ( brute force as you call it ). They will use statistical structures, Bayesian nets for example to make good decisions. They will very accurately build up a model of their oppositions play and exploit that.
Poker has incomplete information but that information is incomplete for humans as well so no big deal.

Practically speaking, humans may not be able to beat the best computers now, I don't know ( ask Darse Billings maybe), but if you are saying theoretically, they can NEVER beat the humans, you have failed to identify a single reason for this.

Perhaps you think we have some God given skill to play poker that is not theoretically possible in computer. I am sure you can start a thread in SMP forum and get 100 responses as to why exactly God made poker only playable by humans but allowed Computers to reign supreme over the much more beautiful game of chess. My feeling is that if similar time and money is spent working on computer poker, the programs will have success comparable to chess.

D.

[AlienBoy]

I wasn't meaning never so much as talking out loud as to "can it" and if so, how?


Bayesian nets? hmmm..


Still, a computer is limited to IF/THEN more or less.

[ QUOTE ]
They will very accurately build up a model of their oppositions play and exploit that.

[/ QUOTE ]

But if they are doing that, because of the ability of a player to manipulate information, the baysian net could be altered, the same way spam still flows through my baysian spam filter.



AB

[David Sklansky]

"Perhaps you think we have some God given skill to play poker that is not theoretically possible in computer. I am sure you can start a thread in SMP forum and get 100 responses as to why exactly God made poker only playable by humans but allowed Computers to reign supreme over the much more beautiful game of chess. My feeling is that if similar time and money is spent working on computer poker, the programs will have success comparable to chess"

Head up games yes. Not ring games or tournaments. Reason being again that the algorithms that beat the best players in chess work just fine against lesser players. Switch to poker though and the computer will be leaving a lot of money on the table against suckers while human experts are extracting more. And trying to fix that problem in a non exploitable way would be almost impossible.

[FortunaMaximus]

Good point. Also the funniest thing I've read in awhile.

I mean, where is the AI going to get the money from in the first place? It has no emotional investment in the outcome and its profits will always come from deficient opponents.

[David Steele]

Switch to poker though and the computer will be leaving a lot of money on the table against suckers while human experts are extracting more. And trying to fix that problem in a non exploitable way would be almost impossible.


If( heads_up){

____If ( opponent = expert ) game_theory_method;
____If ( Opponent = retard ) special_nonoptimal_method;
}

If(multiway){

___If (num_expert_opponents >0)game_theory_method;
___else special_nonoptimal_method
}

You can pretty much tell who the bad players are just from
poker tracker type data, so it is easy to model your opponents for above.

The experts dont play too many pots so this will cover a lot of cases. May take just a bit more work to handle more optimally the case with both types of opponents in the same hand.

Now take the same millions of dollars spent on chess software research and the details can be fleshed out.

D.

[George Rice]

[ QUOTE ]
Switch to poker though and the computer will be leaving a lot of money on the table against suckers while human experts are extracting more. And trying to fix that problem in a non exploitable way would be almost impossible.


If( heads_up){

____If ( opponent = expert ) game_theory_method;
____If ( Opponent = retard ) special_nonoptimal_method;
}

If(multiway){

___If (num_expert_opponents >0)game_theory_method;
___else special_nonoptimal_method
}

You can pretty much tell who the bad players are just from
poker tracker type data, so it is easy to model your opponents for above.

The experts dont play too many pots so this will cover a lot of cases. May take just a bit more work to handle more optimally the case with both types of opponents in the same hand.

Now take the same millions of dollars spent on chess software research and the details can be fleshed out.

D.

[/ QUOTE ]

You are talking about a situation where the experts don't know their playing a computer, and the computer is playing known players online. If the experts knew, David's last sentence addresses your post.

[ QUOTE ]
And trying to fix that problem in a non exploitable way would be almost impossible.

[/ QUOTE ]

The experts would simply play to make the computer think it was playing a fish and then exploit it.

[David Steele]

The experts would simply play to make the computer think it was playing a fish and then exploit it.

Fish play bad all the time, not that "simple" to fake inexpensively. The computer can treat any borderline players as experts and use game theory, while cleaning out the fish with special modules.