Wednesday, June 7, 2017

Facing the Jerome Gambit: Artificial Non-Intelligence

Programmers who create chess-playing engines usually try to make them as strong as possible. They then develop ways to scale back the strength, creating levels of play that can give a human opponent a chance to have a decent game. How, then, to make a smart computer program weaker? This question comes up a few times in the following game, as Bill wall takes on a program at a playable level.

Wall, Bill - GarboChess JS 6.0

1. e4 e5 2. Nf3 Nc6 3. Bc4 Bc5 4. Bxf7+

4...Kxf7 5.Nxe5+ Nxe5 6.d4 Qh4

This move is likely "book" for the program.


There are 150 games with this position in The Database.


The computer saves an endangered piece, but the move is hardly consistent with the counterattacking ideas of the line. It is not surprising that this is the only game in The Database with the move.

8.dxe5 Qxe4 9.a3 Bc5

Keeping the Bishop on an active square, but the program "overlooks" a tactic. The safer move was 9...Be7 with Black still better.

10.e6+ dxe6 11.Qh5+ g6

Kicking the enemy Queen with the pawn is all-too-human an idea; better was probably 11...Qg6. Black has frittered away its advantage with the last three moves.

12.Qxc5 c6 13.Nc3 Qxc2

Pawn-grabbing is a known failing of early and weaker chess programs. Lack of development and King safety are more important issues.

14.Re1 Nf6 15.Bg5 Nd5 16.Qd4 Bd7

17.Nxd5 cxd5 18.Qf6+ Kg8 19.Bh6 Qxb2 

Putting off checkmate for a very short time.

20.Qxb2 Kf7 21.Qg7+ Ke8 22.Bg5 Rf8 23.Qe7 checkmate

No comments: