Learning conceptual chess for testing evolutionary programming versus a reasoning-based soft expert system: The KASER

A soft expert system is one that is qualitatively fuzzy. In this paper, we present such a system known as the ldquoknowledge amplification by structural expert randomizationrdquo system or KASER. This system facilitates reasoning using a domain specific expert and commonsense knowledge. It accomplishes this through object-classed predicates and an associated inference engine. The KASER addresses the high cost associated with the bottleneck of knowledge acquisition. Further, it also enables the entry of a basis of rules and provides for the automatic extension of that basis through domain symmetries. We will demonstrate the learning features of the KASER by comparing its capabilities with an evolutionary programming system that tries to learn the game of chess. In this paper, we concentrate on the evolutionary chess player and also describe the learning capabilities of the KASER, found through other tests. While this EP system may be able to play chess, the KASER provides knowledge as to why certain moves are employed as it learns the game. This powerful characteristic allows the KASER to learn supra-linearly, rather than through exhaustive searches. Thus, the KASER can be applied for many other scenarios in which learning through knowledge acquisition is employed.