The same as 10 years ago
Posted: Mon Oct 01, 2012 11:00 am
When I played 10 years ago on a Pentium 200 MHz against engines like
AnMon and Comet, most of my wins would come by using larger structures of fixed (blocked) pawns.
To my surprise, when I play nowadays on a 4 core machine against engines like Houdini and Stockfish, I am still able to win a fair number of games (although much less and altogether few and far between) by resorting to such pawn structures. Maybe 70% of all winning games would come through this.
Of course, I am happy when I am able to win a game, but at the same time I am a bit annoyed, because the fact certainly also points to a less of progress in engine development for an entire decade concerning this specific element of play. How could this actually happen?
My personal view is that nowadays top engines (Houdini, for example) are at the 2700 level positionally, not worse, but not better. This will make 150 points below the best humans.
At the same time top engines are some 500 elo points above the best humans tactically, Carlsen and Kasparov for that reason. This is due mainly to the fact that humans have difficulties visualizing all relevant moves with all piece movements having taken place, maybe not on move 3 or 5, but certainly at move 7 and deeper. There are just too many piece and pawn movements taking place for the human brain.
Surprisingly, the advantage of engines disappears with larger fixed structures of pawns and humans still may enjoy advantage there. I see 2 main reasons for that:
- one is that programmers, although aware of the issue, are reluctant to tackle it, because their testing ground consists mainly of engines and they consider the casual games won by humans in this way as just some kind of chance events.
- and the second one is that the main engine weapons (attacks and mobility) fail them with larger fixed structures on the board.
With bigger fixed structures attacks occur much less frequently, and what concerns mobility, it is simply not the same as in usual situations - with fixed structures less mobile pieces can gradually but forcefully gain mobility, while more mobile ones gradually but forcefully become useless, something difficult to track by the engines moves ahead.
Of course, it is not a big problem that engines lose some games even when multi-headed, but the annoying thing, at least for me, is that engines generally have not made a considerable progress for an entire decade!, while they are much better in almost any other situation on the board. The problem might not appear in engine-engine matches, but it is disgusting to a point to see an engine player 300 elo points above the best humans lose like a novice, and not a single game at that.
I think it is about time that authors took to solving this issue, which could lead to a much improved positional game.
I would dare to propose an easy solution without resorting to subtleties for trying to at least partly eliminate the problem. I think that assigning bonus points for the most advanced pawns of a group of diagonally connected pawns (assigning bonus points in terms of the size of the group, in terms of ranks and files, in terms of fixed structures, especially important, and in terms of closeness to the enemy king, also very important) could be the cure without implementing much additional knowledge - I do not know how many engines do that. At the same time attacks and mobility could be weighted down a bit in relation to the growing number of pairs of fixed pawns, especially when part of a single continuous structure.
Another point that possibly fails the engines is that they do not seem to consider a blocking knight when part of a group of a diagonally connected pawns with fixed structures as an asset, but it certainly is. Such a knight retains good mobility, cannot be attacked by enemy pawns and in a way makes the structure of pawns whole, because the best the adversary can do is try to exchange that knight. But engines seem to think that the side with the knight just has a deficient pawn structure.
I am sorry if I said something out of place, I just wanted to be of little help and hopefully a day will come when humans will not be able to exploit such engine weaknesses.
It is interesting to know what you think of this.
Best regards,
Ludmil
AnMon and Comet, most of my wins would come by using larger structures of fixed (blocked) pawns.
To my surprise, when I play nowadays on a 4 core machine against engines like Houdini and Stockfish, I am still able to win a fair number of games (although much less and altogether few and far between) by resorting to such pawn structures. Maybe 70% of all winning games would come through this.
Of course, I am happy when I am able to win a game, but at the same time I am a bit annoyed, because the fact certainly also points to a less of progress in engine development for an entire decade concerning this specific element of play. How could this actually happen?
My personal view is that nowadays top engines (Houdini, for example) are at the 2700 level positionally, not worse, but not better. This will make 150 points below the best humans.
At the same time top engines are some 500 elo points above the best humans tactically, Carlsen and Kasparov for that reason. This is due mainly to the fact that humans have difficulties visualizing all relevant moves with all piece movements having taken place, maybe not on move 3 or 5, but certainly at move 7 and deeper. There are just too many piece and pawn movements taking place for the human brain.
Surprisingly, the advantage of engines disappears with larger fixed structures of pawns and humans still may enjoy advantage there. I see 2 main reasons for that:
- one is that programmers, although aware of the issue, are reluctant to tackle it, because their testing ground consists mainly of engines and they consider the casual games won by humans in this way as just some kind of chance events.
- and the second one is that the main engine weapons (attacks and mobility) fail them with larger fixed structures on the board.
With bigger fixed structures attacks occur much less frequently, and what concerns mobility, it is simply not the same as in usual situations - with fixed structures less mobile pieces can gradually but forcefully gain mobility, while more mobile ones gradually but forcefully become useless, something difficult to track by the engines moves ahead.
Of course, it is not a big problem that engines lose some games even when multi-headed, but the annoying thing, at least for me, is that engines generally have not made a considerable progress for an entire decade!, while they are much better in almost any other situation on the board. The problem might not appear in engine-engine matches, but it is disgusting to a point to see an engine player 300 elo points above the best humans lose like a novice, and not a single game at that.
I think it is about time that authors took to solving this issue, which could lead to a much improved positional game.
I would dare to propose an easy solution without resorting to subtleties for trying to at least partly eliminate the problem. I think that assigning bonus points for the most advanced pawns of a group of diagonally connected pawns (assigning bonus points in terms of the size of the group, in terms of ranks and files, in terms of fixed structures, especially important, and in terms of closeness to the enemy king, also very important) could be the cure without implementing much additional knowledge - I do not know how many engines do that. At the same time attacks and mobility could be weighted down a bit in relation to the growing number of pairs of fixed pawns, especially when part of a single continuous structure.
Another point that possibly fails the engines is that they do not seem to consider a blocking knight when part of a group of a diagonally connected pawns with fixed structures as an asset, but it certainly is. Such a knight retains good mobility, cannot be attacked by enemy pawns and in a way makes the structure of pawns whole, because the best the adversary can do is try to exchange that knight. But engines seem to think that the side with the knight just has a deficient pawn structure.
I am sorry if I said something out of place, I just wanted to be of little help and hopefully a day will come when humans will not be able to exploit such engine weaknesses.
It is interesting to know what you think of this.
Best regards,
Ludmil