Using Heuristic-Search Based Engines for Estimating Human Skill at Chess (by Matej Guid; Ivan Bratko; Ljubljana, Slovenia)

Using Heuristic-Search Based Engines for Estimating Human Skill at Chess (by Matej Guid; Ivan Bratko; Ljubljana, Slovenia) 

Interessante o presente trabalho, o qual trata das diferenças quanto a avaliação das posições proporcionadas por diferentes engines. O trabalho é abrangente, todavia dentre as diversas considerações apresentadas, destaco a seguinte:

"Capablanca-Alekhine, World Chess Championship match (game 27, position after Black’s 37th move), Buenos Aires 1927. White is winning after 38. Kf1-e2!,

for example: 38. ... Qc1xb2+ 39. Ke2-f3 Qb2-b3+ (or 39. Qb2-c3+ 40. Kf3-g4!) 40. Kf3-f2! and the white King safely returns to h2, avoiding any further checks. However, Capablanca played 38. Kf1-f2?? and after 38. ...  Qc1-d2+! had to agree to a draw, since there is no escape from perpetual check now. How much should White be “punished” for his mistake? The table on the right shows backed-up heuristic evaluations obtained by various chess programs, when evaluating the moves 38. Kf1-e2 (computer’s choice) and 57. Kf1-f2 (move played) using 12-ply search. While all engines agree that the position is equal after 38. Kf1-f2, there are vast differences among the evaluations of 38. Kf1-e2."