Programs designed for especially good results against humans

[JensBNielsen]

In this previous topic http://www.talkchess.com/forum/viewtopi ... ight=human I made a poor headline: "programs with a human playing style".

I hope this one is better!


Computers may play better against humans in this way:

1) Among almost equal good moves, choose those that lead to open and tactical positions instead of closed and positional positions.

2) Among almost equal good moves, choose those with traps that may lead your opponent to fail.
If you are in a lost position anyway, you may even choose a bad move if it gives your opponent a risk to fail.
This also applies to a draw position, if the bad move still can hold the draw.

3) Use the opponents time to prepare an immediate answer to fx 5 or 6 of his best moves.
This means the program will only search one halfmove less ahead and be a little bit weaker, but it will stress the human, make him tired and give him no time to ponder while the computer calculates its move.

Has anyone tried to implement such features like these?

Jens

[James Constance]

Let's face it, computers don't need any extra help!

Hiarcs has a swindle UCI option http://www.hiarcs.com/pc_uci_options.htm

Doesn't Shredder have a prefer open positions option - or maybe that is Pocket Fritz?

[Ovyron]

And don't forget ChessTiger's Anti-human option, it's specially good against Father like opponents

[Dann Corbit]

And don't forget ChessTiger's Anti-human option, it's specially good against Father like opponents

Crafty also has some anti-human features built into it.
Swindle mode (keeps trying to win in a known draw hoping for a mistake)
Prevention of long pawn chains (traditional anticomputer strategy)
Special code for D00 Stonewall (blitz attack strategy that often works)

I think that there may be other features like this.

[JensBNielsen]

Thanks for your replies.

I assume that noone so far has made a program that replies instantly.

When I wrote the preparation material (80 pages) for GM Bent Larsen and other danish players before their meeting with Baby Deep Blue in Copenhagen 1993, I really feared the machine had such a feature that might totally confuse and stress the players.

Jens

[Mike S.]

Nowadays, all engines are capable of such "instant reply", or almost instant reply playing. On new computers, engines can reach 8 or 9 plies (or even more, + powerful extensions) in less than one second.

As for the antihuman programs, Chess System Tal is another important example.

[JensBNielsen]

But this is not enough if the program is playing against a human GM at 40 moves in two hours.

But if the GM has spent his 3 minutes on a move, the program can prepare an answer for 6 moves using 30 seconds for each move - and that is better.

[Mike S.]

But if the GM has spent his 3 minutes on a move, the program can prepare an answer for 6 moves using 30 seconds for each move - and that is better.

Yes, that is an old and good idea for practical training too, the "multiple" permanent brain. The SciSys (Saitek) Turbo 16 K had it, and I think the 24 K too.

But I am afraid if an engine really has the chance to play against a GM in an important, public match with 40/2h or the like, the programmers will always go for the "conservative" concept. They will probably say, if I let ponder for 6 different moves, almost 5/6 of the ponder time is wasted

Although, it seems to me it would be something very similar to the multi-pv analysis mode, and I would like to see this alternative pondering as an option, at least. But since it's such an old idea which has not been used again yet, I do not really expect it, unfortunately.

[JensBNielsen]

If the program ponders on the wrong move, all the time is lost.

The cost for calculating responses to 6 moves is only one ply as each iteration takes 5-6 times longer than the previous one.

And a program plays almost as well fx at searching to a depth of 13 ply as searching to a depth of 14 ply.

I hope to see a match between such a program and a human some day.

Jens

[bob]

Thanks for your replies.

I assume that noone so far has made a program that replies instantly.

When I wrote the preparation material (80 pages) for GM Bent Larsen and other danish players before their meeting with Baby Deep Blue in Copenhagen 1993, I really feared the machine had such a feature that might totally confuse and stress the players.

Jens

I did something similar in Cray Blitz when playing blitz games against opponents, and you are correct. With the program moving _very_ quickly, the humans (GM level) tended to fall into the same tempo, where they were more likely to make tactical errors even though they had plenty of time on their clock...