Human Computer Chess

Certainly getting a computer to learn chess by memorising previous games isn't really the way to go. There are far too many combinations for this to be useful.

Humans do not learn this way, but they are far better with patterns than computers are. How does a human learn it is best to open with central pawns? Knights before bishops? Castling early? How does a human after more experience learn when it is safe to break these rules to improve position?

Experienced human chess players also can see positional advantage much easier than a computer. It is this ability that grandmasters have to exploit to gain the upper hand over the computer.

Once you get a computer to learn the same way a human does, I think you are mostly there (along with a HUGE amount of memory, of course).

The only way to solve the game of chess is by backward induction so the opening poster is correct in some sense. The only thing preventing it is technology, as due to the enormous amount of possibilities computers are not fast enough and the database storage would be insanely big.

Endgame tablebases are the perfect example of the game being solved by backward induction. It's just going to take a while.

This, however, would not lead to computer's playing more human chess. We are not able to play using backward induction perfectly unless we're relatively close to the endgame.

Is it possible to introduce the principle of evolution in the chess programming tech?

Like coding the rules, defining a goal, starting a big bunch of computers playing eachother, only let the winning computers survive, and the losing to die, two good computers mating (in some way) and multiply?

If the chess playing property was good for survival, then the good old biological evolution did the trick on a few billion of years to produce a Magnus Carlsen. I think computers can do it quicker.

The problem is - how to program?

Originally posted by FabianFnas
Is it possible to introduce the principle of evolution in the chess programming tech?

Like coding the rules, defining a goal, starting a big bunch of computers playing eachother, only let the winning computers survive, and the losing to die, two good computers mating (in some way) and multiply?

If the chess playing property was good for survival, t ...[text shortened]... oduce a Magnus Carlsen. I think computers can do it quicker.

The problem is - how to program?

I can't imagine how it can be done. If it is possible, they might "evolve" their own goals and lose interest in chess.

They might then go on a terminator style rampage and kill us all off for giving them such mundane tasks. 🙂

Originally posted by lausey
I can't imagine how it can be done. If it is possible, they might "evolve" their own goals and lose interest in chess.

They might then go on a terminator style rampage and kill us all off for giving them such mundane tasks. 🙂

I've seen evolutionary programs in action. How to perfect the geometry of an antenna, how to get better wingprofile on an aeroplane, etc. The result can be very surprising, as in biological evolution.

No, I don't think a evolving chess program can berserk by itself if you hardcode the objective, and the rules. We don't have to be afraid for something worse than it didn't produce any good chess program.

Originally posted by FabianFnas
Is it possible to introduce the principle of evolution in the chess programming tech?

Like coding the rules, defining a goal, starting a big bunch of computers playing eachother, only let the winning computers survive, and the losing to die, two good computers mating (in some way) and multiply?

If the chess playing property was good for survival, t ...[text shortened]... oduce a Magnus Carlsen. I think computers can do it quicker.

The problem is - how to program?

A quick search for evolutionary algorithm and/or genetic algorithm will demonstrate that evolutionary principles are indeed used in modern AI. Combine that with a neural net and you might eventually produce something that can learn to play chess by experience.

The simply solution is HELPMATES!

The object of the game is checkmate. It is not too bothered how
it gets there.

So...


You get the box to find all the helpmates in 2 from here.
(that is Black to play and get itself mated in 2 moves).
There are 64! helpmates from the above postion.

If Black plays 1...e5 There is only one mate in two moves from here.


White to play - only one assisted mate in 2 moves.
2.Qh5 Ke7 3.Qxe5 mate.


So you see the machine is already learning. After 1.e4 e5 is the best move.

You then program it to find all the helpmates in 4 then 5 and then 6 etc.
Up to move 40. Chess is Solved.

Originally posted by FabianFnas
Is it possible to introduce the principle of evolution in the chess programming tech?

Like coding the rules, defining a goal, starting a big bunch of computers playing eachother, only let the winning computers survive, and the losing to die, two good computers mating (in some way) and multiply?

If the chess playing property was good for survival, t ...[text shortened]... oduce a Magnus Carlsen. I think computers can do it quicker.

The problem is - how to program?

This already exists for chess.

One of the issues is that you have to pick what criteria are adaptable (or which ones the algorithm considers) and how they mutate as they are deemed more or less fit.

Found the one example I know of (I'm sure there's more out there):
http://www.top-5000.nl/ps/A%20Self-Learning%20Evolutionary%20Chess%20Program.pdf

Originally posted by Palynka
Found the one example I know of (I'm sure there's more out there):
http://www.top-5000.nl/ps/A%20Self-Learning%20Evolutionary%20Chess%20Program.pdf

Cool. Very interesting. 🙂

Originally posted by Exuma
How about this stupid simple idea. How hard is it to program a computer to play chess, but only giving it the rules. Make it so that when it gets mated that is bad. Give it a storage database, so that it can learn from its mistakes, but only in the "I got mated, therefore I must try something else" sense. Or I suppose, the "I checkmated the other guy so th ...[text shortened]... lose, win, learn something. Would this lead to a computer that played more "human" chess?

It is extremely easy. Someone once built a device that did this with nothing but matchboxes and paper notes that, after playing for a while, became unbeatable at tic-tac-toe. I don't remember who, but the details might be in Hofstadter's work somewhere, or in that massive games tome by Conway &al.
Note that I wrote easy. I didn't write "practically possible within the lifetime of an entire dynasty of you and your descendants unto the forty-second generation". It's conceptually very simple indeed... it's just that for tic-tac-toe, you need a smallish stack of matchboxes, while for chess, you need an Earth full of them.

Richard

Originally posted by Palynka
Found the one example I know of (I'm sure there's more out there):
http://www.top-5000.nl/ps/A%20Self-Learning%20Evolutionary%20Chess%20Program.pdf

Wow! Thanks for posting this, I am surprised at how well it works...