The Promise of Massive Parallelism for AI

Innovative Applications of Massive Parallelism

Volume

Papers from the 1993 AAAI Spring Symposium

Contents

To those of us playing with parallel models of AI, the question is often raised as to whether the parallelism offers new ways of doing things, or just lets us do the old ones fast. The implication is that if the latter, the work is just "engineering hacks" and not "scientifically interesting." We won’t mention where this question typicMly arises from 1, but we do know that answering it is sometimes difficult. In this talk, I will attempt to face this question head on. In short, my suggestion is that the answer to "Whether parallelism adds something new OR is it just faster" is "YES." That is, that these two are not crucially different, and that parallelism both lets us explore new approaches and to do traditional techniques faster. However, and more importantly, I will argue that the latter (faster) will provide some the former (new).