ICGA Journal: The Journal of the Computer Games Community

The question “Can Machines Think?” has always intrigued researchers and philosophers. Last year the question revived when Deep Blue played Kasparov, but Kasparov then held his ground. However, recently Deep Blue defeated the World Champion, and so the question became really acute.

This article attempts to define a suitable sword-in-the-stone test, equally valid for computers and humanity. The following tests are successively discussed: the Chess Test, the Turing Test, and the Gödel Test. The Chess Test is considered to be obsolete after Kasparov’s defeat. The Turing Test is illustrated by providing background on the Loebner Prize Competitions. Its restricted version is a problematic sword-in-the-stone test; its unrestricted version is certainly not a test that any machine is going to pass any time in the foreseeable future. Finally, the Gödel Test is discussed.

The evaluation function of a software version of the Deep Thought chess program is tuned against a database of master-level games. The agreement between the move chosen by the program and by a human master is maximized. A set of algorithms solving the maximization problem is presented. The trade-off between computational speed and the chance of getting stuck in a local maximum is extensively investigated.

The performance of the chess program with the tuned evaluation function is experimentally measured. The program has played several matches of 500 games each with different settings against a fixed program. The results show that 98% of the performance of the evaluation function can be achieved in about a week by tuning the evaluation function from scratch against the database. If the evaluation tuning is started from the best evaluation used so far, the new evaluation function improves the playing strength by 34 rating points.

When Deep Blue played Garry Kasparov in February 1996 and May 1997, the extensive IBM Web pages devoted to the site claimed that Deep Blue did not use artificial intelligence (AI). I argue that this claim is (1) inaccurate, (2) representative of a wide-spread phenomenon in the field, and (3) ultimately harmful to AI.