Massively Parallel Retrograde Endgame Analysis
Lewis Stiller · ICGA Journal · 1988
This is a clear example of reasoning by the program.The KPK domain is treated in Chapter 6.Previously, results were published in Advances in Computer Chess 3.Although very simple, the KPK endgame contains a surprising degree of complexity.According to the author: "Although almost trivial to a master, [the KPK endgame] is too complex for any but the most scrappy and vague rules to have been articulated in existing chess manuals"Various successful attempts of catching this endgame into attributes and rules have been made in the past, for example by Bramer.Therefore, this domain serves as a test to check whether the developed method will yield equally good and brain-compatible results.The problem is decomposed into 5 subproblems, all independent, and 31 unique attributes are defined.An example of each subproblem is given.At the end of the chapter, a cost-analysis is presented in terms of programming time and run-time.Finally, in Chapter 7, Shapiro faces the challenge of solving the very complex KP(a7)KR endgame by induction of rules.For this particular endgame, no human codification seems to have been achieved at all.Again, the problem is broken up into (8) subproblems and (36 unique) attributes are defined.In this case however, the subproblems make up a tree-like structure of variable depth.As in the KPK endgame, each subproblem is outlined and illustrated by examples.The program was developed in four stages, each taking about 3-4 man-weeks of effort.Using self-commenting as a diagnostic aid, the system was tuned to yield a complete and correct rule of solution which is both machine-executable and brain-compatible.