Computer Chess by Webgogs

The Delayed Gratification Child

The computer is not the poster child for delayed gratification. An example of the DGC is a child who decides to eat a candy bar later. The computer organizes its moves according to what it thinks are "better than my opponents worst move" hierarchy. Often more than one move rates identical and the computer must rank them. What it does is chooses the ones that it thinks must be reponded to directly. But centuries of chess have taught us that it is sometimes better to build a position before executing it. I do not imply that a computer will automatically execute a tactic when a build up is better. I refer only to positions where the choices appear equal. Such is the situation that is coming up again in Chess.com VS Cheater-1.

I've spent the last 7 years studying computer chess intensely. What i found is that across the board, all programs have this problem. When I got involved in this game against Cheater-1 I knew from the outset that we'd need a long range plan comprised of several turning points where this particular condition would be featured. I had 3 at the ready. I can only talk about this now because Cheater-1 is reading our blog and could have avoided the first one. This would have made it very tough to avoid a draw. Now it's too late for him to change course because to do so means accepting an inferior position now. Not doing so means accepting an inferior position later thus giving him a very slim chance to find a draw. He would like you to believe that he is being beaten by computers. But, have a look at the game. What you see is a limited number of options that most expert players can easily calculate out. So why is he having such a hard time since he has state of the art software on a very powerfull machine? Simply this. We held off eating our chocolate bar. He's up material because his computer is very greedy but he's down big time on position. Does this mean computers can't determine the true value of long range positions? Yes! This is true when the game goes through more than 1 even set of moves to get to positions that are over the computers calculation horizon. Now, if you throw in an exchange sacrifice in there then the computer has an even harder time evaluating the results.

So what happens when there are 2 instances of delaying execution with equal options and an exchange sac? Well, you get a game like ours. All the processors say it's so close that it will likely be a draw. Not one engine sees the impending disaster! If you examine the 14.f4 line in the Dragon where White has castled long you will see a very similar problem. The programs just don't see the impending problems. One reason for this is that computers do not evaluate situations where an opponent falls back then resurges very accurately. Humans also share this problem. One reason is the sheer complexity and number of calulations required and the other is the game demands moves that computer don't like. Computers simply don't like moves that are not directly connected to a logical consequence.

I warned Cheater of this flaw in the computers at the start of this game and suggested that our plan would require that he himself make judgements for the computer. I said our plan would require that he be as good a player as me if he wanted to stay in the game. He thinks that because his computer can calulate 50 moves in 12 hours that he should just trust it more than his own judgement. Well, where is the strategy? Ya a computer can calculate 50 moves of direct logical connection but what happens when a set of options is blurry?

Look again at the position. It seems that we have the advantage despite his material. His computer says that it is equal because we will have to waste time to get that material back. But what happens if we don't want to get the material back. What happens if when he tries to use that material it benifits us to wait before we take it back? Would you go pawn chasing if you thought you could Queen? What happens if we ratchet down the position even more then go take material back? Hmmm I wonder...

I must confess. I handed out analysis with 1 forged line and 1 missing line in it. I did this because there was a real possibility that Cheater wouls obtain a copy. Oh how he would squeal with delight at the prospect of having the inside track. That would allow him to jump ahead and begin the search for a draw. So I employed a little propaganda trick. By the time you read this cheater, it will be too late. You see cheater, I know that even a modest computer would be a problem to beat in such a simple position. But Cheater, I've been lying on the blog about how drawish this was right from the get go. I didn't want you to have the slightes idea what was up!

Win or draw I'll do a write up of this game complete with extensive analysis when it's over. Here's what I know. Any attempt other than sticking to the draw lines by Black opens the door for a win by White. The line with Qb6 is a try for advantage. A century of empirical knowledge has not been overturned. Black must be very wary of seizing the initiative in the opening!