If you followed a bit the news last month then you likely heard that a new milestone was reached in artificial intelligence. The computerprogram AlphaGo defeated in a match worldclass-player Lee Sedol in the boardgame go with the large margin of 4 -1. The most astonishing of this result is that the program used contrary to his colleagues of other boardgames. multiply times an algoritme based on pattern-recognition built up via self-tuition of mastergames.
The current top-engines in chess are using very advanced algoritmes which can calculate many moves ahead. However this way we can't solve chess in the nearby future. It is no surprise that many amateurs wonder if we can't learn from AlphaGo to ameliorate our chess-engines. At least 1 programmer already tried it: Matthew Lai. He developed the program Giraffe which succeeded by self-tuition in 72 hours to obtain the level of international master (see deep learning machine teaches itself chess in 72 hours plays at international master).
2400 elo must be considered fantastic but at the same time also poor. A.f.a.i.k. nobody managed before to write a program which learned to play chess autonomously by many hundreds (thousand) rating points and above all in just 72 hours. On the other hand an engine of 2400 elo can't compete at all with e.g. Stockfish and Komodo.
I don't doubt further improvements are possible with the path chosen for Giraffe but it is a total other thing to create a new number 1. Personally I believe pattern-recognition is less useful in chess than for go. Our current best engines show every day that brute force is in most cases sufficient to solve a position. In the past we witnessed many times that extra intelligence in our engines (e.g. pattern-recognition) will just deteriorate the strength of a program.
Chess is a very exact game in which the smallest difference in a position can create a total different solution. An example of this butterfly effect was already shown in my article einstellung effect but the most beautiful examples are of course found in the world of problems. Such problems/ studies are also often called twins. Most occur in helpmates (by coincidence Chessbase published recently some) but also in orthodox problems we sometimes find them as in below cute example.
|Mate in 2|
b) Shift Qh7 to a7
c) Shift Ke6 to c6 from postion b
d) Shift Ke4 to c4 from position c
Of course this does not mean that recognizing patterns is useless for chessplayers. We are after all no engines. Contrary as every experienced player will be able to recognize a large amount of patterns of which he hopes to benefit from. I had this luck in my recent interclubgame against Rob Michiels. Rob deviated intentionally from theory but anyway we got a position on the board which I had seen before.
Solution Mate in 2 (Theme Allumwandlung)
a) 1. f8(B), Kf6 2. Qf5#
b) 1. f8(R), Kd6 2. Rf6#
c) 1. f8(Q), Kb5 2.Qfc5#
d) 1. f8(N), Kd6 2.Qc5#