Computer architectures for artificial intelligence processing

Abstract

Since the conventional numerically oriented von Neumann computers are not appropriate for artificial intelligence (AI) software, new architectures for this application are discussed. AI computational requirements are considered. The requirements as well as the architectures for a set of AI applications are presented. Lisp architectures including functional programming and the SPUR (symbolic processing using RISCs) system as an example are described. Logic oriented architectures are considered including OR-parallelism and AND-parallelism. Knowledge oriented architectures and neural networks are discussed.