Chess is an interesting activity for a variety of reasons. “Missed Opportunities” is a new series on things that are sometimes missed during games. When the clock is ticking, humans often have trouble concentrating.

Our previous column and Chess position drew plenty of interest. It was pointed out that Stockfish couldn’t find this move, wanting to play 1. … Kg5. This is a good move, but not clearly winning. In computer science, we call this the horizon problem. Basically, since the algorithmic method is looking ahead move-by-move it can’t see the winning strategy beyond its maximum ply-depth.

We had the pleasure of talking to the AlphaZero development team at the London Chess Conference recently. AlphaZero uses a neural network approach to learn pattern recognition, and maps this into a rule-set database. This is a lot more a human approach to Chess. AlphaZero looks at fewer Chess positions that the brute-force approach of something like Stockfish, but looks at those positions more intelligently. Like humans, it is better at “pruning the tree” of possible good moves.

This was best illustrated in the last match between Gary Kasparov and Deep Blue. Everyone proclaimed machine had beaten human since Gary lost the match. However, the game that Gary won was far more interesting. He won because of a small pawn weakness that he exploited. It was far too subtle, and too many plies ahead for Deep Blue to evaluate correctly, no matter how many processors or threads were added to Deep Blues computational array. [Deep Blue would “clock” long before it would find it.]

Pattern recognition obviously makes human thinking more efficient and superior, but it can lead to interesting moments of what we call “Chess Blindness.” AlphaZero as it has learned from playing Chess has had these moments too. Luckily it stuffs these blunders into its database immediately. [It’s a quick learner!] Learning from defeat is one of the life skills CIC teaches in its programs.