mclure@SRI-UNIX@sri-unix (12/08/82)
This experiment was produced a few years ago and recently updated. It attempts to determine the "relative roles played by tactics and levers according to a player's rating." Tactics consist of the usual pins, forks, and combinations all players are familiar with. Levers are the Kmoch-designated "subclass of positional moves involving two opposing pawns capable of capturing each other", thus entailing a sort of pawn structure consideration (improving pawn structure, decimating opponent's pawn structure, etc), the basis of positional play. (For the record, I haven't read any of Kmoch's writings.) The experiment consists of 24 positions of which 12 are tactical and 12 are positional. The subjects (rated from 1000 to 2435) were given 2 minutes per move. Each can write a "preferred move". Spaces are provided for 2nd, 3rd, and 4th choices for partial credit. "Scoring is done by 1/N where N goes from one to four as the choice-preference of the correct move(s)." So a correct preferred move is worth 1 point. If the 2nd choice is the correct move, 1/2 pts. 3rd choice is worth 1/3, and 4th choice is worth 1/4. It seems to me that the fundamental problem with this experiment as regards human subjects, perhaps even a fatal flaw, is the idea of levers. For example, I scored 3.5 out of the 12 tactical positions, corresponding to a tactical rating range of 1600-1799, which is probably correct for me; however, I scored 9 out of the 12 lever positions, corresponding to a positional rating range of 2200-2399, clearly ridiculous. The only subjects who scored 9 or better on the lever positions were rated 2300, 2360, and two IM's rated 2400 and 2435. Merely knowing there is either a tactical or a positional solution enables one to inflate the positional score enormously. It is fairly easy to determine if a given position calls for a tactical or positional solution. In the former case, laborious calculations are required to verify that the combination exists. In the latter, a fairly quick scan of the pawn structure suffices. The experiment was also run with a number of computer subjects. In the case of the humans, the balance of tactics to levers was fairly even with masters and beyond. Below that, tactics predominated slightly and the scores on the tactical positions were usually higher. In the cases where positional predominated, the balance was still close. With the computers, the tactical value was usually much higher, indicating an extreme bias toward tactical play, a well-known facet of computer play although subjective until this test.