LatestChess - Articles - How Your Chess Program Defeats You

The history of computer chess is long and fascinating, and it has had a profound influence on the field of Artificial Intelligence, though I will not go into that history here. Under the assumption that it is wise to understand your enemy, this article will explain, if not all the nuances, at least the fundamentals of how your computer program regularly defeats you in a game of chess. Let us refer to our common digital enemy as the satanic 'Chessifer'

Introduction: Trees and Master Forest Rangers

I assume that you are not a computer geek, or you would probably already know how your computer chess program works and would therefore not be reading this. Since you are not a geek, I will write this so anyone can understand it without specialized geek training.

The history of computer chess is long and fascinating, and it has had a profound influence on the field of Artificial Intelligence, though I will not go into that history here. There is an outstanding online lecture on that topic that I referenced in an earlier blog post, if you would like to know more.

Under the assumption that it is wise to understand your enemy, this artilce will explain, if not all the nuances, at least the fundamentals of how your computer program regularly defeats you in a game of chess. Let us refer to our common digital enemy as the satanic 'Chessifer'

It is an oversimplification, but not totally inaccurate to say that Chessifer considers each move on the board, thinks of all your likeliest replies, then his replies again, and so on as far into the future as time and processing power permits. At the end of this process he evaluates the resulting positions, decides which is the best and then makes the move that gets him closer to that outcome.

If you think that is not unlike how you play chess, you would be correct, only Chessifer does it much faster and much further into the future than you possibly can. If you have read Kotov's book "Think Like a Grandmaster", you are aware that he advocates imagining a 'tree' of possibilities as you think ahead. That is, the current position is regarded as the root of the tree. Your potential moves are branches that emanate out from this root, and your opponent's replies to your moves branch outward from each of those. The leaves of the tree represent the farthest positions that you can imagine with your necessarily limited brain power.

This kind of a chess game tree is precisely how computers represent the game. Figure 1 is an abstract visualization of a tiny fraction of this tree. The top square or 'root' of this tree represents the initial position before any moves have been made. All of the possible moves by white are shown as branches from this root that lead to the next board position from which Black plays. The label on each square, or node, of the tree indicates the move that was played to reach it. But the node itself actually represents the entire board position after that move is played. We follow the convention in computer science where trees grow downward, but don't let this confuse you.

Figure 1: Partial Chess Tree

As I said above, Chessifer builds as much of this tree as he can, and when he reaches a leaf node (any node that does not have branches to any lower nodes), he uses a program referred to as the 'evaluation function' but which we will refer to as 'Eval' and assigns a numeric score to that position. The higher the score, the more valuable that position is deemed to be for Chessifer. Eval is the part of the program that encodes actual chess knowledge. Thus it will award points to a position based on material, mobility of pieces, pawn structure, and so forth. Programmers work with chess masters, and if a chess master can explain in concrete terms why a given position is better for one side or the other, then this knowledge can be encoded into Eval.