The Singularity Is Here In Chess
Chess machines can already beat all humans
Kenneth Regan, the same as our Ken, is besides many things, an international chess master. He combines his strong understanding of theory, of computing, and of chess. This combination is almost unique in the world and has led him to work on many interesting questions concerning chess.
Today I am proud to announce that Ken’s work on chess is highlighted in the New York Times Tuesday Science section.
Players can use computers to cheat at chess today—even at the world championship level. It is believed that programs that run on laptops can play way above all human players today, and the gap is growing as the programs get better and the laptops get faster and have more memory. A player was even caught red-handed having consulted a program on his smartphone during the final round of last June’s German Championship.
Of course tournament officials are always on guard to see if anything is amiss. But some cheating is alleged to have escaped human surveillance. This is where Ken enters the fray.
Ken’s work started as a method to indirectly discover whether or not a player cheated by using a computer. The central idea was simple: look to see if there was more agreement with the players moves and the program than would be statistically reasonable. But it is not simple. Comparing human play with machine play raised many questions that went well beyond the narrow goal of detecting cheating.
What Ken and his co-authors have done in the last few years is study the record of chess games, not with the goal of just cheating detection, but to understand better how humans make complex decisions under pressure. This work is discussed in the Times article and of course in technical papers that Ken has written with his co-authors Guy Haworth and Giuseppe DiFatta and Bartlomiej Macieja. Here are the two most recent papers:
- K. Regan and G. Haworth, “Intrinsic Chess Ratings.” Proceedings of AAAI 2011, San Francisco, August 2011.
- K. Regan, B. Macieja and G. Haworth, “Understanding Distributions of Chess Performances,” to appear in the proceedings of the 13th ICGA conference on Advances in Computer Games, Tilburg, Netherlands, November 2011.
I will just summarize what Ken has done by quoting him: I have created a statistical model of decision making in chess. As input the model takes (a) deep-and-broad computer analysis of chess positions, and (b) parameters modeling the skill level and attributes of a human (or cyber) player. It outputs (c) estimated probabilities for every move in every position by such a player, and (d) projected error bars on associated statistics, which appear to be within a factor of 1.15 (best-move choice frequency) to 1.4 (error frequency) of “true” in field tests done to date. The former error-bar test was first reported in our blog here. This work is based on over 20 million pages (figuring 2K/page) of data on human decision-making under pressure. It is significant that it’s all from actual competition, not a simulation where there can be nagging doubt about true incentive as in some economics/social-choice studies. Ken and student helpers have written Perl code to collate the data and C++ code to analyze it, almost 200 dense single-spaced pages of code.
In my opinion Ken’s work is quite non-trivial and could have applications to other human decision problems. In any event Ken has been working on it for the last several years. And now his work is good enough to be used in actual cheating cases and more. This is in the Times article. Way to go Ken.
Will a time come soon when we will “cheat” in doing mathematics by using a computer?