Chess vs. Hold'em: Repost

(edited repost from below)


I believe expert level chess has already been programmed with the

Deeper Blue experiment. Expert level hold'em has not.


With chess it's just a matter of figuring out what is the best move based on

future moves against a single opponent..For example if I have 10 possible moves.

I simulate move #1. Then I would have to simulate my next 10 possible moves:

move #1a, #1b, etc to get the best possible outcome. This would enable me to think

many moves ahead. I'm sure there's even more to it, like implementing and

recognizing strategies.


Now think about all the complexities of hold'em.

(their are some preflop. But not too deep).

How many players, called preflop? How many bets preflop (1,2,3, capped?)..

Now the flop..Should I bet my ace high against a 589 flop against 1, 2, 4, or 8

opponents? What about KT2? (And 2-3 hundred more flop textures/hand types)

Is the leader to my left or right? Is the leader

aggressive or passive. We're already getting deep and we're only on the flop.


Now the turn...We have a board of KT29, but how does the turn card relate if at all

to not only the previous flop action, but previous preflop action? Is it a flush

card? or did it pair the board?


What about what my opponent thinks of me? Or What my opponent thinks I think he thinks about me?


Keep in mind that you have to have nodes branching out for each level, each flop type,

etc..By the time you reached the river you would possibly have hundred's of

thousands of decision nodes. Some of which may possible have multiple actions.


It could possible take a team of 20+ expert hold'em player/programmers a few

years to accomplish such a feat.


If anybody really doesn't grasp what I am saying, they simply haven't thought

it out much..Next time your playing think about all the variables you would

need by the time you reach the river, and how long it would take to input them.


You also could buy Turbo Hold'em and laugh at how bad it plays, but also appreciate

the time it must have taken Bob Wilson to program.


Then take Bot Author who is a self-proclaimed amateur who hasn't even spent too

much time on his bot....He is without a doubt full of it that his program

can beat 2/4 in the long run. It doesn't add up.

In (almost?) all of the Alberta groups papers are reasons for why they study poker - things that separates poker from many other games, and especially games that previously have been studied in AI, of which the prime example is chess. If we contrast poker with chess, poker have multiple individually competing players, incomplete (hidden) information regarding the game state, uncertain information (bluffing, style), need for opponent modelling and adaptation - chess have only two competing players, complete (public) information of the game state, certain information, no real need for opponent modelling and adaptation. This is off of the top of my head, so I might have forgotten something (I'm tired too), but it should be enough to illustrate the differences. Especially the need to understand and adapt to your opponents is hard for an AI program, or "bot," since, at least if it is supposed to play against human opposition, it must understand human psychology to some degree. Of course some of these things are not that important against really weak human opposition, though. And humans can play poker extremely bad.


Chess can be played well by a computer by brute force calculations - just get a powerful enough computer. Poker needs much more sofistication. The pros (of AI) haven't gotten this right yet. So probably no amateur has either. If you have a bot, you need pretty weak opposition.


So poker is harder than chess for a computer. Now that does not mean that it can't be the other way around for humans. What comes easy for computers often comes hard for humans, and the other way around. If this actually is the case in the chess/poker case, I do not knaw, as I do not play chess (at any level of skill).


Anyone who want to know more on the state of the art when it comes to poker AI, go to the Alberta group's page and read their papers. There is a link on the Favorite Links page at this site (The Computer Poker Page). You can play against these state-of-the-art bots if you want to too. Should you fear them? Of course they are no complete suckers. But they are actually pretty unsofisticated in their "thinking," especially the opponent modelling part.

First of all Chess was hard!

Go back and look at the Deep Blue crew PHD thesis and papers if you don't agree.

It is not just brute force that was used.


Some hints that it was hard:


- Deep and inovative heuristics

- Custom hardware in Deep Blue case

- extensive preperation with world class players.

- 30 years of University research in many locations

- extensive corporate commercial funding

- treated as the holy grail of Comp Sci.


Now take Holdem Bots for the Internet poker


- There is only now an incentive to create them.


- Holdem skill is not some black magic


- They don't have to play like world class players to win a lot of money


- even the old Lokibot on IRC won pretty easy.

Chess got easier by stronger computers is the argument. The 30 years of research allowed programmers to realize that an AI algorithm wasn't going to beat humans at chess, only brute force calculating programs would. (In fact Deep Blue is NOT strong enough to consistently play at the World Champion level, Garry Kasparov made mistakes in that match that beginners would not have made).

Everyone knows there are two fundamental elements of poker: statistics and psychology. Computers are fantastic at statistics, but they are horrible at psychology.


In fact, computer programming that tries to emulate psychology ultimately collapses back into statistics. For example, a programmer might have the computer recognize that a certain player has bluffed on the river 20% of the time, and the computer would modify its plays accordingly. That's not psychology, but statistics.


At basic poker, statistics is probably more important than psychology. In a 1/2 hold'em game, a beginning player can win simply by playing good starting hands (the one's that are statistically likely to hold up), and making statistically sound plays post-flop.


At the most advanced levels of poker (big bet hold'em or a World Series final table, for example), psychology is probably 90% of the game.


For this reason, I don't think a computer will ever be programmed that can beat the world's best poker players. Those players will be able to get a good read on the computer by looking at its playing patterns, but the computer will never be able to master the psychology necessary to read the players.


My $.02.

Comparing hold-em's complexity to the game of chess is ridiculous. The number of possibilities in a hold-em hand is enormously smaller.

I totally agree that DB was not WC level, I also know the match games and really thing IBM chickened out by not playing a sufficient number of games before dismantling the machine.


It is not fair to say it is just "brute force", it is very complex stuff. Also you can consider the PC based machines that are very strong with much weaker hardware.


As for AI, it is traditional to remove anything from the field that actually works, neural nets, machine vision etc. The tree searching algorithms used in the chess programs are essentially the

outgrowth of work that started in the earliest AI

departments. Even the DB team has roots in the famous AI university department at Carnegie Mellon.


Chess was hard but you don't need to be nearly as good with your Poker pot. Also I think poker is much easier to program then chess, its just that its totally different and not much public work is being done on it.


D.

"..Comparing hold-em's complexity to the game of chess is ridiculous. The number of possibilities in a hold-em hand is enormously smaller.."


On what calculations do you base that absurd statement? With 9 or 10 players X 52 potential flopped cards in many MILLIONS of potential combinations to be made with 2 card hands at the table, and all the various millions of potential player action combinations based on the above variables, how can you possibly make such an absurd, "OTOH" statement??


Calculations please.

learn a little bit about decision trees and exponential expansion