Chess experiment

[CRoberson]

I've tried several experiments with weak players and assistants. The following seemed to produce the best learning experience.

The weaker player has an assistant and the assistant has a computer. On each move, the weak player decides his move for himself but he is allowed to ask the assistant is that move a blunder - notice I said allowed not forced - forced comes later. The assistant is only allowed to reply yes or no. The weaker player decides on another move when he gets a yes and plays the move when he gets a no.

This seems to work well for players below 1500.

Anybody ever tried that? If so, how did it go? If not, try it and tell me your results.

[towforce]

I've tried several experiments with weak players and assistants. The following seemed to produce the best learning experience.

The weaker player has an assistant and the assistant has a computer. On each move, the weak player decides his move for himself but he is allowed to ask the assistant is that move a blunder - notice I said allowed not forced - forced comes later. The assistant is only allowed to reply yes or no. The weaker player decides on another move when he gets a yes and plays the move when he gets a no.

This seems to work well for players below 1500.

Anybody ever tried that? If so, how did it go? If not, try it and tell me your results.

If that's what the process, what's the assistant for? The computer could do that.

[CRoberson]

[
If that's what the process, what's the assistant for? The computer could do that.

To teach human players.

[Frank Quisinsky]

Hi Charles,

that my first thoughts as I read the Google news about learning, before LCO was available.
We can try to produced the style of Petrosjan, Spassky, Fischer, Aljechin, Karpow and Co with learning of games.

Over 3.000 Kortchnoi games are available for us.
Enough material!!

I think it make more sense to simulate playing styles of grandmasters with KI.
Much more interesting as to produced 100 NNUE engines, available with 10.000 different compiles.

Best
Frank

Computer programs are our slaves.
It seems that we are the slaves from our computers.
We can used KI with much more intellgence or the computers are thinking we are stupid!

[Frank Quisinsky]

That is really interesting.

Spassky - Fischer, 1972
Bishop takes h2 (Reykjavik) ...

So each one of us can beat Bobby Fischer in such a position.
Not in reality but GM are humens only!

Means if KI works fine all is possible!
And to have a chess computer with a very realistic style from our 16 World Champions and other players would be a great idea.

Your move chess & games have a lot to do with all the orders.
Sure that people in Papua Neu-Guinea will order such a chess computer.
Should run with battery and 100Mhz is enough for the playing strength and simulation.

And we can say ...
We are using KI for such a simply order!

[towforce]

Over 3.000 Kortchnoi games are available for us. Enough material!!

No. 3000 games at, say, 50 moves per game is a sample size of 50*3000 = 1.5e5 positions. I've never seen an estimate of the number of possible positions lower than 10^40. So 3000 games doesn't tell us enough about how he would have played in at least 10^35 positions. I think that training an NN to play in his style would fail. Worst of all, Korchnoi's own NN would have changed considerably over those 3000 games.

What you'd need to do is to find a way to modify an EF to play in his style. Some chess porograms have tried to do this.