Written by David Tebbutt, MicroScope 11/85 - scanned


Psst! Want to know what the hot topic for 1986 will be? Expert systems, that's what. We've got our spreadsheets, databases, wordprocessing and even outlining/idea-processing products well under control. We've added the widdly bits like keyboard macros and Sidekick. Those of us so inclined have started communicating and we're all ready for the next opportunity to crank value from our PCs. Or, in your case, to crank revenue from your customer base. But what the heck is an expert system and how will you know when one is staring you in the face?

Here's an example. One expert system sits by the production line at a cigarette factory. Whenever the machine starts mangling cigarettes, the operator describes the state of the cigarette and the expert system recommends remedial actions in the form of adjustments to the equipment. If the evidence is inconclusive, the expert system may ask a few questions in order to clarity its views before suggesting causes and remedies.

It is important for the operator to regard the expert system as an advisor or assistant because it lacks two things which we humans have more or less in abundance: general knowledge and common sense. If a rat had somehow entered the machine and started chewing the cigarette, it is unlikely that the expert system would have been given the knowledge to anticipate such an eventuality. Expert systems can be applied only to very thin specific application areas.

Other examples might be health risk assessment ("on the evidence, Mr Tebbutt, you are a prime candidate for a nervous breakdown"), fault-finding in electronic equipment or monitoring and controlling a chemical process. The number of applications is limited only by your imagination and by what types of problem expert systems are good at handling.

Since it is possible to examine an expert system's line of reasoning, it can also make a very good teaching aid ("Give Mr Tebbutt a course of psychotherapy. Why? Because..."). Students can mess around with example problems, examine the expert system's reasoning, and get a good feel for a new knowledge area.

Where these systems make a unique contribution is in non-mathematical reasoning. They really can work with words and with sentences, albeit carefully structured ones. A piece of expert system knowledge might look like, this: IF the expert is willing to cooperate AND the problem is well defined AND normal programming methods are inappropriate THEN consider building an expert system.

Rules such as these would be loaded into a 'knowledge base' by the expert or by a 'knowledge engineer' someone who is skilled at extracting such rules of thumb from an expert.

Questions are also built into the knowledge base in order to enable the system to extract information from the ultimate user of the system. A question might be "Is the expert ... willing ... unwilling ... don't know". Some systems prefer to ask "how willing is the expert?" and expect a percentage or score by way of reply.

No two expert systems I've seen are alike. Some are easy to use while others are real dogs. Some try to assess degrees of certainty, others prefer to take a non-mathematical approach.

Most systems expect the user or knowledge engineer to load rules in the form described above. Interesting exceptions to this spring from work by Professor Donald Michie and his team. His products work out their own rules from examples given to the system. You can enter the examples by hand as with most other systems but, more importantly, some of his programs can ferret through a file of data, extracting rules from that.

Large expert systems applications require a lot of human effort in order to build a knowledge base from scratch. With automatic rule-induction from a database (Peter Laurie's Superfile has this option) this hurdle can be overcome.

Most expert systems today are only a shadow of what is to come. Many people, especially those in corporations, are buying expert systems to see what they can do. This way, when the machines and the software become more powerful, the companies will be well positioned to take advantage of them. Having said that, many companies who bought expert systems to play around with ended up building really useful applications.

For those prepared to invest a little money and effort, expert systems may turn out to be another source of added-value revenue. Not only could you sell expert system 'shells' (empty of knowledge) but you could also consider training and supplying 'knowledge engineers' to help your customers exploit their new purchases.

Some expert-system companies are willing to give or sell 'run-time' versions of their products. If you have access to expert knowledge for which there is strong local demand, you could consider building a knowledge base and selling it with a run-time module.

The opportunities are there. It's early days yet but give it a couple of years and everyone will be trying to jump on this bandwagon. By then you could be an established purveyor of expert systems.