Distributional analysis of large data-sets of chess games played by humans and those played by computers shows the following differences in preferences and performance: (1) The average error per move scales uniformly higher the more advantage is enjoyed by either side, with the effect much sharper for humans than computers; (2) For almost any degree of advantage or disadvantage, a human player has a significant 2–3% lower scoring expectation if it is his/her turn to move, than when the opponent is to move; the effect is nearly absent for computers. (3) Humans prefer to drive games into positions with fewer reasonable options and earlier resolutions, even when playing as human-computer freestyle tandems. The question of whether the phenomenon (1) owes more to human perception of relative value, akin to phenomena documented by Kahneman and Tversky, or to rational risk-taking in unbalanced situations, is also addressed. Other regularities of human and computer performances are described with implications for decision-agent domains outside chess.
[1]
Ivan Bratko,et al.
Using Heuristic-Search Based Engines for Estimating Human Skill at Chess
,
2011,
J. Int. Comput. Games Assoc..
[2]
A. Tversky,et al.
The framing of decisions and the psychology of choice.
,
1981,
Science.
[3]
Ivan Bratko,et al.
Computer Analysis of World Chess Champions
,
2006,
J. Int. Comput. Games Assoc..
[4]
Ivan Bratko,et al.
HOW TRUSTWORTHY IS CRAFTY’S ANALYSIS OF WORLD CHESS CHAMPIONS?
,
2008
.
[5]
Guy Haworth,et al.
Understanding Distributions of Chess Performances
,
2011,
ACG.
[6]
Guy Haworth,et al.
Intrinsic Chess Ratings
,
2011,
AAAI.
[7]
M. Glickman.
Parameter Estimation in Large Dynamic Paired Comparison Experiments
,
1999
.