Printer Friendly

Expert systems expertise.

While I may be old enough to pass for an expert, I seem to have less qualifications for that title as time goes by. I go to conferences, technical exhibitions, and company visitations. I even read a lot. But I keep coming up against developments that I know precious little about.

Last year at a robotics conference in Sweden a learned speaker was asked if a situation he'd described in his paper couldn't be best handled by an expert system. His response was that the didn't know enough about expert systems to say aye or nay.

That brought me up short with the realization that I didn't either. I vowed to learn. So I was pleased to find that the 1985 American Society of Mechanical Engineers' International Computers in Engineering Conference was loaded with presentations on expert systems. I went. They had over 500 preregistered for the conference. Including those people who came only to the accompanying exhibition, there were over 100 in attendance.

Since that is a small percent of the readership of this magazine, I feel a strong obligation to tell the rest of you something you missed. Take the following in the context of the sure knowledge that one cannot become an expert on expert systems in three-and-a-half days at a conference, no matter how fast he takes notes.

It is well publicized that computers don't make mistakes. All computer error is the fault of the people who design, program, or operate the inanimate object. It is also well known that computers don't think. They crunch numbers and process words, but they create neither on their own. Computers can be built with huge memories and can search and sort what is in them at fantastic speeds. They can print more reports in an afternoon than you can read in a week. That is on the practical side.

On the hype side, there is still an attraction to calling computers smart, teachable, intelligent, knowledgeable, and expert. People reluctant to use those human attributes in describing machines counter this ploy with cries of definitions of these terms. The best answers seem to be performance oriented. The idea is to make the machine act as if it were intelligent, etc.

Suppose you have 70 engineers designing power plants and all must at one time or other specify valves to go into the subsystems they are designing. Each engineer has experience that can be indexed as IF--THEN statements. If certain conditions prevail, then certain components are applicable. Some of the engineers have valve application knowledge the others don't have. If you could put all the IF--THEN logic statements from all your engineers into a big enough computer, and program the machine to search and sort them when a set of design conditions are presented, the machine would identify which valve in all your catalogs in the memory fills the bill for that application. Wow, that's an expert system. It could be held that such a system exhibits more expertise than any one of your engineers alone.

We have just described a knowledge representation and search method of the rule-based variety. They also come in frame-based and logic-based types.

Such a system could tell you, on the basis of an analysis of the oil around a transformer, that you better shut it down and change the oil, or it is likely to blow up all over you. Maybe you want to pick end-effectors or grippers for your robot's arm. Anything where conditions suggest actions or conclusions.

These still fit the old concept of garbage in--garbage out. If you don't have access to enough expertise to load valid knowledge into the system, there is no way that the system is going to come up with valid choices. Likewise, if your rules call for inappropriate actions your machine won't know the difference. That is not to belittle the systems. They are really an advancement over the old method of trying to remember everything you've learned about a given field. It was reported that one structural expert said his whole career had been reduced to 50 rules.

Some of these systems also are described as "transparent." This means that after they print out the answer to your problem, you can question how the conclusion was reached and the machine will go backward through the steps and tell you what conditions led to the conclusion. They also are called flexible in that additional IF--THEN statements can be added as more experience is recorded, and others can be deleted if later shown inappropriate. Wouldn't it be easy to say that such a system is learning by experience?

One of the biggest attractions to the idea is the certain knowledge that in many fields the old experts are retiring.
COPYRIGHT 1985 Nelson Publishing
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1985 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:comment on International Computers in Engineering Conference
Author:Keebler, Jim
Publication:Tooling & Production
Date:Oct 1, 1985
Previous Article:Machine-tool builders cope with change.
Next Article:Carpenter matched tool and die steels.

Related Articles
Companies and organizations indicating an interest in bidding for the OPTN and SRTR contracts.
ADVISORY/Experts Available to Comment on The Future of the Skyscraper.
Experts Available to Discuss Warning for DNS Cache Poisoning Attacks.
Systems engineering; proceedings.
Experts Available to Discuss FERC Plans for New Security Regulations for Energy Companies.
Experts Available to Discuss Cyber-Security Standards Approved for Electric Industry.
Eventscom- Bahrain CEO & Founder to Share Event Marketing & PR Expertise.
Bayanat Airports Out in Force at Airport Show 2008.
Engineering of complex computer systems; proceedings.

Terms of use | Copyright © 2017 Farlex, Inc. | Feedback | For webmasters