Machine intelligence: an examination of two rival philosophies-knowledge-based expert systems vs. neural networks

摘要

Examines two approaches to machine intelligence: knowledge-based expert systems and neural networks. The framework for the analysis of both approaches is the implementation of two functional systems, each of which can be used independently to solve identical problems. The first system, the hack mechanic's helper (HMH), is an automated advisor designed for car repair, which was developed using Texas Instruments' PC Plus, a knowledge-based expert system shell coded in LISP. The second system, mecha-net, is also an automated advisor for car repair, but its implementation is achieved by neural network software developed using the C programming language. The strengths, weaknesses, and limitations of both approaches to machine intelligence are exposed by solving the identical problem and then comparing the systems' performances.