Printer Friendly

Epilogue.

In our introduction to this special section we stated that although the results of the Fifth Generation project do not measure up to the expectations it generated, nevertheless those involved with the project have a sense of achievement. What is the source of this discrepancy between the generally negative perception of the project and the generally positive feeling of the people who actually participated in it?

Perhaps the essence of this contradiction lies in the difference between the way the project was presented initially to the public and what the project really was about. The promoters of the project in Japan popularized the project by promising to make the dream of artificial intelligence (AI) come true. This view was further amplified by scientists throughout the world, who capitalized on the fear of Japanese technological supremacy in order to scare their own governments into funding research.

However, what the project was really about was evident very early to anyone who cared to find out. Ten years ago, one of us stated in this publication:

The smoke cleared when ICOT was formed, with Fuchi as its director. With the excuse of budget constraints, all ballasts were dropped, and a clear, coherent research project emerged: to build parallel computers, whose machine language was based on Horn-clause predicate logic and to interface them to database machines, whose data-description and query language was based on Horn-clause logic.

The fancy artificial intelligence applications of the original proposal remain, serving as the pillar of fire that gives the true justification for building faster and better computers; but no one at ICOT deludes himself that in 10 years they will solve all the basic problems of artificial intelligence...Commun. ACM 26, 9 (Sept. 1983), 637-641.

Technically, the project can be said to have achieved the first part of the objective: several parallel computers have been built. These computers, collectively known as Parallel Inference Machines (PIMs), incorporate the programming language KL1, which is a concurrent logic programming language based on Horn-clause logic. KL1 was used to implement PIMOS, the PIM operating system, as well as to build various applications that exploit the parallel processing capabilities of the PIMs. Moreover, the PIM machines can be considered to achieve a peak performance approaching one "gigalips" (one billion logical inferences per second), which was one of the main tangible goals of the project (and hardly thinkable at the time it was first proposed).

The second part of the objective, to build database machines, has been less clearly addressed. It was reformulated, as a result of the perceived success of the concurrent logic programming paradigm, into building a software-based implementation of logic data-bases in KL1 on the same PIM hardware platform.

As for the AI applications, the project successfully demonstrated the PIM technology on some interesting and novel parallel applications in various areas of mainly nonnumeric computation, including AI.

If, then, in keeping with Fuchi's vision, the project has indeed built a bridge between parallel computing and advanced applications including AI, as is certainly arguable, why the mixed feelings?

First, 10 years ago many believed that within a decade parallel computers would dominate, and the key strategic question would be: who has the best parallel computing system? The Fifth Generation project certainly aimed at offering its sponsoring companies a strategic advantage in that regard. The reality is that parallel computing is still a niche market, due to the incredible increase in the power of ordinary workstations and high-end PCs, and due to the difficulties that still hamper the programming of parallel computers. The current application of parallel computers is confined mostly to large, regular, numeric computations, which are simple enough to implement with reasonable effort in a concurrent dialect of a conventional language such as Fortran. The demand for parallel computers to run a wider class of computations, such as those which in practice require programming in higher-level languages, does not exist yet. Consequently, the technological success of ICOT in building complete parallel computing systems aimed at wider and more difficult-to-program tasks did not resonate in the computer industry.

Second, many believed that the AI boom of the early 1980s would continue indefinitely. Had this happened, everyone would be begging by this time for more computing power to implement their fancy new AI algorithms. The technology developed by the Fifth Generation project could have found a more ready market in such a case. The reality is that the results of AI research of the past 10 years are disappointing, at least compared to the expectations many had a decade ago. There are few, if any, researchers in AI who find themselves limited by the computing power available on the fastest workstations, and there are few, if any, exciting AI applications that would be even more exciting if they had more computing power at their disposal.

Hence ICOT had some difficulties in identifying interesting AI applications to implement on the PIM machines, and although the applications chosen demonstrate the potential of ICOT's parallel computing technology, there is not a huge market for advanced AI applications demanding such technology.

Third, it seems there is still a big gap between ICOT's parallel kernel language, KL1, and the AI applications the Fifth Generation project set out to address. The concurrent logic languages, such as KL1, lack some of the key attributes of logic programming--its declarative nature and built-in search capability--that were present in the original logic programming language, Prolog, and which helped to motivate Fuchi's original choice of logic programming as a suitable bridge between AI and parallel computers.

Some researchers believe a better way to realize Fuchi's vision is through systems that exploit parallelism automatically in Prolog or other logic programming languages which preserve its essential attributes. The progress made in this direction is encouraging--it even seems possible to combine the attributes of concurrent and traditional logic languages.

In Summary, it can be said that ICOT has built a bridge between parallel computers and AI applications. However, with the two ends of the bridge being (perhaps temporarily) out of favor, and the bridge itself being weaker than it might be, it is perhaps too soon to expect the inauguration of the bridge to be greeted with great acclaim.
COPYRIGHT 1993 Association for Computing Machinery, Inc.
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1993 Gale, Cengage Learning. All rights reserved.

 
Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:Technical; The 5th Generation Project: Personal Perspectives
Author:Shapiro, Ehud; Warren, David H.D.
Publication:Communications of the ACM
Article Type:Cover Story
Date:Mar 1, 1993
Words:1032
Previous Article:Evan Tick.
Next Article:Editorial pointers.
Topics:

Terms of use | Privacy policy | Copyright © 2018 Farlex, Inc. | Feedback | For webmasters