A New Use for Monte Carlo

It may be interesting to communicate that the origin of the “Monte-Carlo” feature in some RYBKA software had something to do with the matches described in Kaufman (2010). Shortly after joining the RYBKA team, I wanted to refine the basic piece values used, as they seemed to me seriously off-base. I had the idea that if I could play thousands of hyperspeed computer games starting from positions with various material imbalances, I could judge from the results whether the evaluation function was mis-scoring those positions. I asked Vas Rajlich if he could make a RYBKA version for me that would make this possible, and he did so. Strictly speaking it was not truly “random” or “Monte-Carlo” as it was designed never to repeat a game, but the details are not important here. Suffice it to say that for my purpose it was effectively sufficiently random so as to insure plenty of variety in the games without degrading the level of play too much.