Computers playing chess

[tel0004]

Today me and my uncle had a debate about computers playing chess. I was asking why its possible to beat a computer at chess. Im talking about deep blue and all that stuff. He said humans have analytical thinking which computers dont have, but I said that isnt needed in chess. Chess is purly mathmatical. Now if two computers with the same programing played each other, whoever when first should win 100% of the time, NO exceptions. Right? So by that theory, if a human played an absolute perfect game (which I dont think is possible) then he should be able to beat a computer, but if he makes one mistake, the computer should have the edge and should win. I was also thinking that if the computer went first, it should be impossible to win. Obviously chess computers arnt perfect, which they should be, we have had 40 years to get it right, so thats why they can be beat. So why is it possible to beat a computer, they can think 50 moves ahead. This would make an interesting dissertation, too bad im a finance major.

 

[TeddyKGB]

I don't know if even a supercomputer can store all the total move probabilities of every possible board layout.

 

[tel0004]

No, it wouldnt store the possibilities, but use complex math equations.

 

[Nightson]

I'm still waiting for the computer that can play go well.

 

[Helo]

The problem with computers it they cant recognize bluffs and traps and the like. We have creative thinking that computers dont

 

[TeddyKGB]

No, it wouldnt store the possibilities, but use complex math equations.

"Complex math equations" is a trifle vague, don't you think?

My point is that the number of possible games far exceeds the number of atoms in the universe. No memory space can hold any more than a miniscule fraction of those, using "complex math equations" or not.

Humans can beat other humans at chess because humans share tendencies. Computers are not innately wise to these tendencies, but then humans probably don't think like computers either. Perhaps that is why the computer vs. human matches have been inconclusive thus far.

 

[Sojourner<><]

Say for instance a computer is able to play an absolutely perfect game (which is for the purposes of this argument a game in which every move is planned from the beginning) and it has the fortune of the first move. In order for the game to work out perfectly, the opponent would be required to also be playing for perfection. If he/it makes one single irrational move, the computer's plan could possibly become imperfect even if the move puts the opponent at a disadvantage.

 

[TeddyKGB]

Say for instance a computer is able to play an absolutely perfect game (which is for the purposes of this argument a game in which every move is planned from the beginning) and it has the fortune of the first move. In order for the game to work out perfectly, the opponent would be required to also be playing for perfection. If he/it makes one single irrational move, the computer's plan could possibly become imperfect even if the move puts the opponent at a disadvantage.

This is why there is no single "perfect game." A particular game may be called "perfect" if each move produces the maximal probability of winning. But which move that happens to be for every board configuration will change depending on what one's opponent does.

 

[Sojourner<><]

Right, but if both players are playing to maximize each move according to the current board configuration, then it would become possible for the player with the first move to effectively predict the move of his opponent on each round, allowing for a complete game plan to be developed beforehand even if each player fully calculates the entire course of the game at each move. This is why I had to establish what I meant by a "perfect game" because in this scenario each player is playing perfectly. Any randomness in these calculations at all would make it an impossibility.

 

[Sojourner<><]

Here's a challenging question...

Say your opponent goes on the offensive and takes a few of your pieces in an obviously planned maneuver that you did not foresee. How would you then be able to logically determine that the maneuver was planned and in how many turns beforehand?

 

[Sojourner<><]

Right, but if both players are playing to maximize each move according to the current board configuration, then it would become possible for the player with the first move to effectively predict the move of his opponent on each round, allowing for a complete game plan to be developed beforehand even if each player fully calculates the entire course of the game at each move. This is why I had to establish what I meant by a "perfect game" because in this scenario each player is playing perfectly. Any randomness in these calculations at all would make it an impossibility.

This is assuming that the best possible move in a given scenario can be determined using a set of rules instead of calculating the probabilities of success.

 

[TeddyKGB]

This is assuming that the best possible move in a given scenario can be determined using a set of rules instead of calculating the probabilities of success.

Hmm. Is there such a rule set that can replace straight probabilities?

 

[Soul Searcher]

I would think if both players played perfectly the only possible outcome would be a draw.

 

[Sojourner<><]

Hmm. Is there such a rule set that can replace straight probabilities?

I don't know of any by I don't see why an algorithm couldn't be developed if we're thinking in terms of the strongest move to strengthen your position without taking into account future moves.

This would just be the first stage... if the opponent is using the same algorithm, then the first player could determine what it's next move is.

I just realized though that this is also assuming the opponent doesn't have the same ability to predict the first player's moves. If it did, I think that would be an infinite amount of calculations.... overflow.

 

[Sojourner<><]

I would think if both players played perfectly the only possible outcome would be a draw.

I think that's true unless one player has the ability to predict the other.

 

[Soul Searcher]

I think that's true unless one player has the ability to predict the other.

To play a perfect game both players have to be able to perdict the other. Still results in a draw.

 

[Sojourner<><]

To play a perfect game both players have to be able to perdict the other. Still results in a draw.

yeah but it would be impossible for both sides to perfectly predict the other's predictions. That's why I had to redefine what I meant by a "perfect game".

 

[Soul Searcher]

yeah but it would be impossible for both sides to perfectly predict the other's predictions. That's why I had to redefine what I meant by a "perfect game".

That is correct but if neither makes a mistake and each makes the perfect move to counter the others previous move then the only possible result would still be a draw.

 

[CSMR]

Now if two computers with the same programing played each other, whoever when first should win 100% of the time, NO exceptions. Right?

If the same program were used in both computers all the time (no random element) then either white would always win or black or it would always be a draw. It isn't necessarily true that white would always win.

So by that theory, if a human played an absolute perfect game (which I dont think is possible) then he should be able to beat a computer, but if he makes one mistake, the computer should have the edge and should win.

By the definition of a mistake yes, provided the player's best game can win.

I was also thinking that if the computer went first, it should be impossible to win.

If it were the best possible program, but that assumes white has a winning strategy - I don't think that is known.

So why is it possible to beat a computer, they can think 50 moves ahead. This would make an interesting dissertation, too bad im a finance major.

They don't do all the analysis for 50 moves ahead; that would be too many possibilities.

 

[FSTDT]

So why is it possible to beat a computer, they can think 50 moves ahead.

There are lots of reasons why a computer can be beaten:
- The capacities of the computer are limited to the skill of the programmer.

- A lot of people think that computer predict moves by brute force, or by calculating every possible orientation of the board. But chess programs dont work like that, otherwise it would take 100 years for the program to deduce in the first move that the most strategic move is b5-e5. Most computers are programmed with predefined (or nearly-predefined) patterns for the first 10 or 15 moves of the game. In fact if you watch two computers play against each other, they'll literally fly through the first 10 or 15 moves, because its practically pre-programmed. Similarly, they are practically pre-programmed for the end-game. Using the brute force method and normal chess time constraints, the most powerful supercomputers only have the capacity to calculate the best sequence of moves up to 12 to 15 moves ahead (this is still better than a humans 10-move upper limit).

- Computers arent programmed for perfect play, but rather for best probability due to the normal time constraints of chess rules. Computers have to be programmed like this, because they do not have an infinite amount of time per move, nor an infinitely fast computing power.

- To some extent, there is a random element involved in most commercial chess software. You can open your pieces in exactly the same way everytime, and a computer may not behave the same way each time. The relative strength of the computers opening moves may vary slightly.


I am a programmer, and I know that it would be pretty impossible to program a computer to analyze every possible board orientation per move, but I believe an unbeatable computer opponent can be made like this: have the computer store the outcome of every game into a database, and program the computer to mimic successful gameplay and avoid unsuccessful gameplay based on the contents of the database (this is the equivalent to "learning"). Allow two different computer opponents to play against each other endlessly, and eventually you'll have (in a digital sense) the laws of natural selection working to produce closer and closer approximations to the ideal computer opponent.