Chess as the Drosophila of artificial intelligence

Making chess the Drosophila of artificial intelligence

Chess and the formalization of concurrent action.

1. The Berliner thesis problem.

2. Consider two pawns vs. a king on a very large board.  We
suppose that the two pawns are on the same rank, and the king
is in an arbitrary position.  It seems that if the two pawns
are far enough apart, one of them will queen assuming that the
side with the pawns is to move.  Formalize this.

A. S. Kronrod once proposed that chess is the Drosophila of
artificial intelligence, i.e. that much of the experimentation
in making machines behave intelligently is appropriately
carried out in the context of chess.

This is a plea to carry out artificial intelligence experiments
in a domain chosen for its technical suitability for experimentation
rather than its utility in itself.  The analogy is that geneticists
did their experimentation on fruit flies (Drosophila) rather than
on elephants.  While it would be more useful to breed better elephants
than to breed better fruit flies, fruit flies go through a generation
in a week, and 1,000 of them can be kept in a bottle.

Chess has the following advantages for artificial intelligence work:

1. Comparison of the performance of a program with human performance
is reasonably straightforward.

2. Many problem positions are available in which it can be
determined whether the program found the right move.

3. Chess involves only some of the intellectual mechanisms
required for general intelligence.  This would be a disadvantage
if chess were proposed as the sole experimental domain.  However,
artificial intelligence has proved to be a difficult science, so
it is important to study intellectual mechanisms separately in
so far as this is possible.

The realization of chess as a domain for AI research has not
proceeded very well.

	In the first place, funding sources have
been over-sensitive to possible charges of wasting money, although,
to my knowledge, there have been no actual such charges by
Proxmire, et. al.  This shyness has extended to the point where
a sponsor of AI research asked that its support of a thesis that
used chess as a domain for research in pattern matching asked
that its support not be acknowledged in the published paper.

	Secondly, and perhaps as a consequence of the first, most
people who write chess programs have sporting motivations (and
more recently commercial motivations) more than scientific
motivations.  Therefore, they often don't write papers, and
success in tournaments is regarded as more important than
understanding the intellectual mechanisms and publishing
their discoveries in the scientific literature.

	We have some proposals for alleviating these problems.