I considered a GA based chess AI for a time, before I realized that there would be no rational way to explain the rook-king endgame. At that point I went online to see what I could find, and realized that chess games generally follow canned openings until the human deviates from the standard opening. Chess computers also use canned end games, which explains away the problem I had with the rook-king end game. The rules used in the rest of the game (protect the king) are abandoned for an end game that requires attacking with the king.

Then I encountered an article that suggests that an AI for chess would be novel, because there is no I in chess, so every I would be AI by definition. The study showed that human chess masters get where they are not by learning how to think through combinations, but by having lots of experience and a very good memory of other games. Studies show that grand masters generally evaluate only one or two moves, they just choose the right moves to evaluate. Other studies showed that if a grand master was shown a position from an actual game they were far better than a novice was at reconstructing the position from memory. However, if the same masters were shown a board where the pieces were positioned at random, then they were no better than novices at reconstructing the position from memory.

The point is that chess computers use a brute force approach to evaluate every possible move, and fancy algorithms to snip off branches that appear unpromising. Meanwhile, human chess masters are simply playing and remembering so many games that they can evaluate a position by subconsiously comparing the position to games in memory.

At that point, I abandoned the idea of a chess AI in favor of building a robot. There is no evidence that either humans or machines use intelligence to win chess games. Instead, the machines use brute force, while humans use encyclopedic memory.