Pachman and Kuhnmund wonder why it is so hard for the computer chess
programs circa 1986 to play chess well.
p.31-32"How is it possible that such a problem [providing a complete
solution to the problem of devising a chess playing machine], by its very nature, can overtax even the best computers which
can guide rockets to the moon and beyond or control the running of a whole industry?"
Pachman and Kuhnmund speculate that efforts to limit the moves examined by computer chess programs
would require the machines to follow a set of rules to search only the moves that are 'meaningful'. Despite the efforts of
the best minds, this strategy had not (1986) produced a machine that could play chess at a master level.
p.83-84"Despite the impossibility of giving a computer the intuitive capacity of a grand master,
many chess programmers have opted for Shannon's B strategy in order to restrict as far as possible the number of positions
which need to be evaluated. This means that a good program must contain instructions for determining the most meaningful moves
in a given position by applying basic precepts arising from our present state of chess knowledge. In the course of time, programs
based on this strategy have led to a clear increase in strength on the part of the chess computer, despite the gloomy predictions
of some experts. However, it has become equally clear that the optimistic forecasts of some [Shannon strategy type] B programmers
that the world would soon witness the advent of a new electronic World Champion have proved equally naive. Despite strenuous
efforts within universities renowned for their pioneer work on chess programming, it has not proved possible to improve the
strength of B strategy programs to the level of master chess."
Perhaps careful selection of the moves to be analyzed will give our machine the intuition of a grandmaster.
But what would those rules be?
p.87"If a B-strategy programmed computer could only acquire the
intuition of a grand master, it would be in a position to outplay any human opponent. The crux of the matter lies in the choice
of key moves to be analyzed, and it is precisely here that the evaluation criteria hitherto used by the machine are, compared
with human methods, far too inexact and unsophisticated."
Grandmaster Hubner says that he 'make[s] various observations about the effectiveness of the pieces'
before he selects his move in a chess game. Perhaps our computer chess program could do that as well.
p.90"[gandmaster Hubner interview with Spiegel newspaper in 1979]Concrete calculation is easily
imitated, but there are other processes taking place in the subconscious mind which are difficult to isolate but which nevertheless
produce identifiable results. I first gain a general picture of the position and make various observations about the effectiveness
of the pieces. Then, on the basis of these elements, I come to a decision about which move to play. The bulk of my observations
never reach the conscious mind. To transfer such human thought processes to the machine would require a complete and quantified
register of strategic concepts"
Once again, we see that machines of the 1986 era were good at finding tactical moves but weak at
finding positional moves. It seems that there (still) is a need for a heuristic to improve the positional play of the machines.
p.131"We can draw two important conclusions from the above examples [a computer chess program attempts
to find the best move in two positions]. Firstly, well programmed computers have already shown that they are highly efficient
when it comes to calculating precise tactical sequences. As we shall see later, however, their strategic capability (choice
of a plan, maneouvering, in positions with few tactical elements) is distinctly lower."
