Human and Machine Representations of Knowledge

Abstract

Four ex1st1ng Knowledge-representations for the computat1on of s1m1lar functions 1n a chess endgame were 1mplemented on the same computer 1n the same language. They are compared w1th respect to effic1ency regard1ng time-space requirements. Three of these programs were then paraphrased 1nto English and all four were studied for their feasibility as 'open book' advice texts for the human beginner in chess. A formally verified set of rules was also tested for its suitability as an advice text. The possible effectiveness of these advice texts in 'closed book' form is considered. The above experiments comprise a case study of a phenomenon known as the "human window". This phenomenon mot1vated an analysis of four documented instances of mismatch between human and machine representations. These are:

Three Mile Island II Air Traffic Control, III NORAD Mil1tary Computer System, IV The Hoogoven Royal Dutch Steel automation failure