Expert Systems in Decision Support

R.B.Jones, 82/11/1


In discussing the exploitation of expert system it is not easy to find a happy medium between an unrealistic overassessment of the possibilities and an unduly conservative approach which more or less misses the point. I believe that most of the ideas I have presented are necessary and possible rather than utopian and unrealistic. Where my suggestions go beyond the capabilities of existing expert systems they do so in pursuit of a distributed expert capability fully integrated with more conventional networked information processing facilities. This is inevitably a problem area which will be of greater interest to commercial than to academic institutions, or at least, which only becomes of interest as expert systems become commercially viable products.

I conclude with a handful of what I consider to be key ideas, a cautionary note, and a heartfelt plea.

Enable algorithmic techniques to be mixed with deductive techniques and don't write a set of rules where a subroutine can be supplied, (appropriate technology)
Enable full access to system resources to be obtained through expert systems (qua intelligent interfaces).
Support distributed knowledge bases.

To these I add my cautionary note:

iv) Chose your applications carefully.

There really are a lot of problems which expert systems will not be able to handle for us for a long time. In particular, still, anything that really requires intelligence. Some guidelines can be laid down for the shcrt term. Amng them I suggest:

Don't expect an expert system to solve problems which your best available human expert can't solve. You might be lucky, but you probably won't be. Someday they may be able to beat us but not yet.
A careful systems analysis is essential, in particular a breakdown of all the information which a human being needs to solve the problem. The expert system won't be able to solve the problem unless it has access to all the same information.
Watch out for some of the intractable problem areas. A relevant one for ICL is understanding programs. If your problem requires you to look at a program listing and understand what it's doing then you're unlikely to get an expert system to solve it for you this century. If you can go about solving your problem in a reasonably systematic way, with maybe a few judgements of a woolly nature that you wouldn't know how to program, then you may be O,K. with an expert systen, but if you have to pace up and down till your brain aches awaiting inspiration, then the problem is all yours. And of course, if the main problems are human problems, persuading people, or understanding them, then there is no hcpe.
And finally for my plea.

The most impressive thing about the Japanese fifth generation report is neither their ambitious objectives, nor the techniques they propose for attaining then. It is the appearance of a substantial collection of papers all working in a well co-ordinated way towards the same overall objective. One is immediately made avare of the fragmentary nature of western research.

By contrast, the Alvey Report, which is more or less Britain's answer to the 'fifth generation' seems a despondent half hearted venture. It explicitly doubts that British Industry will collaborate without being given financial inducements to do so, and even though proposing that government funds be applied to secure this essential collaboration, the proposals comes across as being, by constrast, rather fragmentary.

Notwithstanding the lack of any co-ordinated assault by British Industry there remains a real possibility that ICL, with some collaboration with other companies, could make good progress by itself. To be successful in this ICL needs to mobilise all its best creative talents throughout the company. This should not be viewed as a few new isolated product developments, it is rather, a new approach to the relationship between man and machine which should leave few products untouched.

ICL sells products to meet the information processing requirements of its customers, and in doing so its main activities are themselves processing information. Expert systems techniques are the last element in the jigsaw which provides an information processing facility capable of handling all an organisations information processing requirements with a mininum number of staff dedicated to computerese. They offer not only a new direction for our products, but also a means of substantially improving our own 'information productivity".

To achieve all these wonderful things, it really is quite essential that we all work together, and the most important element in working together is good communications. Notwithstanding the requirements of security, we cannot produce an integrated set of products if we do not know to what goal we are aiming, and we will not lead in the technology of the fifth generation if the parts of the company working on expert systems do not talk to each other or to anyone else.

issue 1/43

up home © RBJ dated 1982/11/1 HTML 1996/6/6 edited 1997/4/18