Chapter 6

Expert systems and their impact on the professions

WR Williams

Bill Williams’ introduction to electronic data processing came in 1962 when on secondment to the Royal Navy. He continued his interest and was eventually appointed to the Defence Department’s then newly formed Computer Division. Retiring from the Navy in 1970 he joined the pioneer software house SPL (now Progeni). He also joined the Computer Society and has served both as chairman of the Wellington Branch and as a national councillor. While still with Progeni he took over the part-time post of secretary. When it became apparent that more support from the national office was needed to cope with the society’s expanding activities he ‘retired’ for a second time so as to be able to devote more time to society affairs. Currently he is serving as executive director.

It is probably true to say that it is members of the professions, lawyers, doctors, architects and educators who, here in New Zealand, have been the slowest to appreciate the effects that the new technology will have on their work. Trade unions were also rather slow at getting off the mark but, once alerted, have taken a very active interest in developments. The effects on skilled manual workers are perhaps more noticeable, after all it could not have been too difficult to see what word processors would do in the office or how robots would change the production line. Even so, one would have expected the professions to be equally perceptive but, while there will hardly be a union official who is not anxiously looking at the impact computers will have on his members, the same cannot be said for the professional bodies.

One major profession has been left out of the list given above, namely that of engineering. This is because, here at least, there is a strong core of practitioners using advanced computer technology in their work. Mention has already been made of the Ministry of Works in this respect. There are also other honourable exceptions and New Zealand is not without its pioneers in the field of expert systems; individuals working in such areas as computer aided learning and specialised areas in medicine and law. But professional bodies as a whole lag behind. There are still graduates coming from New Zealand law faculties and the medical schools knowing less about computers than the average sixth former. Yet all will be using computers within the next decade and hopefully there will be some, as they become more senior in their selected fields, who will be active in the design of expert systems themselves. For the time cannot be far away when expert systems will be universally recognised as the best way to pass on the hard won knowledge and experience of a lifetime’s work.

If we are going to talk about expert systems it would be well to define exactly what we mean. They are a part of the wider subject of ‘knowledge engineering’ which again is a sub-set of artificial intelligence. While the concept of an expert system has been around for some time the term itself is relatively new. For our purposes we will define it as:

A computer application which uses the previously stored knowledge of those highly qualified in a particular field, to solve problems that otherwise would require the human intelligence of similarly qualified persons.

An early example may best illustrate what is meant. Back in the beginning of the 1970s an experiment was conducted in Glasgow to try out the idea of using a computer to take a doctor’s initial case notes. It was surprisingly successful. Most had expected that patients, who had predominately a working class background, would resent answering questions fromwhat was in effect no more than a computer controlled typewriter (no readily available VDU screens then). But this was not the case. The patients said that they were less intimidated by the machine than by the presence of a doctor. They were more inclined to say that they didn’t understand the question, to ask for clarification and to take their time in answering. Quite unexpectedly it appeared that they were also more truthful, almost all admitting to imbibing more alcohol than when questioned by the doctor himself!

Responses for the most part were confined to a simple YES/NO/DON’T KNOW. The doctors who participated had had to sit down and formulate the questions so that they could be answered in this way and to construct a sequence of questions that would take note of previous responses so that, for instance, a patient who had been identified as a male would not later be asked questions relating to childbirth. The programs used were primitive, they probably would not be regarded as an expert system today. Indeed the procedures covered only one facet of expert sys­tems as we know them. However, whether or not this was an expert system it did have the one essential ingredient, that of co-operation between experts and computer system designers. Since then much more sophisticated systems have been developed. Those who remember the BBC programme Now the Chips are Down will recall seeing a team of doctors headed by a well-known specialist working on such a system. The aim was to make available to doctors in all parts of the world the same assistance they would receive if they were able to consult the team personally.

In the latter example we have the essence of expert systems. They are at their best when they have been prepared by specialists in clearly defined areas for use by professional people in their day-to-day work. The specialist almost certainly will need the help of those who have studied and thoroughly understood the advanced computer technology involved but in the end a particular system will stand or fall on the ability of the expert to communicate his knowledge and skill in this way. He, in turn, will do this best if he has a good understanding of what is involved on the technical side.

There are in New Zealand a few who are studying expert system design. One of these, Graham Wrightson of Victoria University, has written a paper to be published in the NZCS Journal on this subject. The present chapter draws heavily on material in that paper. In his introduction Graham Wrightson says:

Electrotechnology has wrought great changes in the way that professions fulfil their various roles. Figure 6.1 offers an estimate of how the four principal components of the professional’s work have shifted in the past and will shift in the future. Physical activity has declined and probably will continue to decline as a result of increasingly powerful, accurate and dexterous electromechanical devices. Over the last century, entrée into professional work has required increasingly long training and apprenticeship, a process catylyzed by the growth of specialization. But knowledge systems should soon begin to assume much of the burden of memorization and information retrieval, leading to a steady decline in the duration of professional training. Service and delivery — the actual output of professionals — have decreased recently in proportion to their other activities. But this decrease, caused by the extra effort required of modern specialists to keep pace with the exponential growth of information, will shortly be reversed as artificial means are developed for assimilating and applying knowledge.

The only major component of a professional’s work that I believe has not changed and will not change substantially is judgement, which might be seen as the constant by which the professions are defined. When judgment vanishes, the profession vanishes. As Figure 6.2 illustrates, the capabilities which comprise professional judgment change over time. Largely in response to the enormous growth of information generated by electrotechnology, professionals today need a balanced mix of capabilities in order to perform effectively. In the future I expect knowledge systems to reduce the time a professional will need to spend in memorization, gathering, analysis, and reasoning to reach useful judgments. However, intuition and perception, two distinctively human skills, will become more and more important, especially in proportion to other skills. Knowledge-based systems, and expert systems in particular, have recently gained significant economic and scientific importance. There are several reasons for this. Firstly, there has been an ever increasing demand for expert consultancy; secondly, there have been tremendous cost reductions following the computerisation of expertise otherwise only available from highly trained specialists, and thirdly, expert systems have accomplished some amazing results.

He goes on to provide examples of expert systems that have proved their worth. These are listed in Table 1. Included in this list are two New Zealand systems, DAMP and COMIX, which Graham Wrightson and his students at Victoria University have helped to design.

Expert systems would not be possible without the computer and, indeed without the very powerful computers of today. Not only must they be capable of holding and searching through large volumes of data and of exploring the well nigh endless permutations and combinations that are inherent in such systems, but they must also be capable of carrying out a meaningful dialogue with the user. Perhaps we could look briefly at some of these factors.

Techniques for holding data and for searching through data files for particular records have been under study for many years. Typically, expert systems have to handle large quantities of poorly structured data and they present some of the more difficult problems designers encounter in this area. None of the methods available could be considered ideal, each has its strengths and weaknesses, and it is a matter of choosing the best to handle each particular application. A correct choice is vital. A mistake at this stage could result in a clumsy or even inoperable system. A particularly disquietening feature here is the fact that the error may only become apparent when attempting to run the system under full operating conditions; trials with smaller test files may well have failed to reveal the inherent weakness. The system will almost certainly have cost several man years of effort by this time and all this will have been wasted.

The next requirement is, however, the heart of the expert system. Graham Wrightson described it as the provision of a ‘domain-specific reasoning system’. It comprises several parts. First there is the part which codes the information supplied by the enquirer (the situation data) into a form that can be matched with that held by the system (the knowledge base). Then there is the knowledge manager, sometimes called the interpreter, inference engine or logic engine, that takes the situation data and identifies that part of the knowledge base which is applicable. It then applies the programmed logic to work out the inferences from the matching pieces of data.

Computers are good at this sort of thing. It is relatively easy to program a computer in the form:

If A is true then B is the consequence/take this action.

I hope none of my readers will have been the recipient of a piece of computer logic that goes something like this:

If account is outstanding for over three months/and previous reminders have been ignored, then we have a bad payer send out a nasty letter.

The problems to be overcome by the authors of expert systems are two-fold and they present each of the partners, the expert and the systems designer, with a formidable challenge. On the systems side the problems can be quite simply stated — size. A computer program can relatively easily test all possible combinations in a game of noughts and crosses and will never lose, but no computer in existence can do the same for a game of chess. Clearly some method has had to be devised to bring processing within reasonable bounds and it is here that most effort in recent years has been directed. The so-called ‘fifth generation’ computers are going to help, indeed, one of the justifications for their development is their use in expert systems. Fifth generation computers are being designed to enhance the problem solving, inference making and man/machine relationships of the computer. Specifically they will handle ‘knowledge’ rather than ‘data’ as do our present-day machines. Most early work on fifth generation computers has been done in Japan and it is confidently expected that they will become commercially available in the early 1990s. We do not, however, have to wait until then, for today’s computers are quite powerful enough to handle many useful applications. Special computer languages such as PROLOG (PROgramming in LOGic) have been developed. These languages are not concerned with manipulating numerical quantities but with expressing relationships. No distinction is made between a piece of factual information (i.e. data) and a piece of program. PROLOG statements may be regarded as constituting both the specification of the program and the program itself.

The contribution of the experts to the overall design is crucial. Somehow or other they must pass on their knowledge in a form that can be used by the programs. This can be a daunting task. It is one thing to be an expert but quite another to be able to articulate one’s expertise in a coherent form, useful to the computer. While every possible circumstance and combination of circumstances that can occur must be envisaged and the consequences worked out in advance, not all logical combinations will be valid considerations and these can be eliminated. Furthermore, facts established in earlier lines of enquiry can eliminate possibilities when considering situations further down the chain. By identifying these cases early in the specification the expert can make the task of the computer designer that much easier and, in the ultimate, make the difference between describing a system capable of being handled by today’s computers and one that can not. Again the consequence may not always be clear cut; there may be a number of possibilities and situations will arise where there are possibilities within possibilities. Having worked all this out the recommended action may not be to deal with the most likely first since failing to act on one of the others could have serious consequences should it prove to represent the actual case. Considerations such as these must be foreseen and also specified in advance.

The third area that we have identified is that of communication between the expert system and the user. The dialogue, and I have used that word deliberately, between the user and the system must be couched as nearly as possible in the everyday technical language of users and must be meaningful to them. Input is usually achieved by the user supplying standard data pertaining to the situation under consideration, followed by a series of questions and answers. The questions are supplied by the system as it identifies the additional information it needs to proceed. Output can vary, depending on the type of system, but typically it comprises either a diagnosis or a recommended line of action or both. There is, however, a further essential ingredient. This comprises an explanation as to why the particular diagnosis or action has been chosen, providing sufficient detail for users to validate the reasoning, at least to the extent of their own knowledge and experience. Not only does this build confidence but it also guards against the possibility of gross error, due either to misleading information having been supplied, or errors in the system logic when faced with unusual circumstances.

The requirement for good communication between system and user will be self-evident but what is not self-evident is the complex programming and the amount of processing that is needed to achieve this end. Here, as in other computer applications, the more ‘user friendly’ the system is, the greater has been the effort that has been put into the system design and the more power will the computer need to run it.

To get a feel for what is involved let us look at designing a simple, very simple expert system — say ‘Instructions for Crossing the Road!’ We must all be more-or-less experts here or we would not be around to tell the tale. Probably the first instruction we received or have given to our children runs something like this:

Look RIGHT

Look LEFT

Look RIGHT again

If there are no approaching vehicles in sight

WALK (NOT RUN) STRAIGHT (NOT DIAGONALLY) ACROSS THE ROAD.

So far so good. But this may not be a very efficient system. For example the user could be waiting unnecessarily if the approaching vehicle is a long way away and travelling slowly. Again, if taken literally, it could be dangerous should a parked vehicle be obscuring the view. We will need to take these factors into consideration. But before we pursue this line of reasoning further, we should reflect a minute for there are other considerations that could make these enquiries redundant. What if there is a traffic officer present controlling the traffic flow or there are traffic lights? Then the cross or not to cross decision becomes a simple matter of whether or not we have been signalled to do so. Well not quite. This probably applies only to the case where the traffic officer is present. It would be safe, if perhaps not strictly legal, to recommend crossing on a quiet Sunday morning where there is not a vehicle in sight for miles around even if the lights say ‘WAIT’. Again, what if there is a pedestrian crossing? Here the opposite is the case. While we may legally be entitled to step on to the crossing it would be imprudent to do so if it appears that an approaching car is travelling too fast and would be unable to stop in time. It would also show commendable courtesy not to insist on one’s rights and force a bus to stop when our saving would be a trivial ten seconds at the most. Further, what if we are not at a pedestrian crossing but one is near? Well the road code gives some guidance here, saying that if we are within 20 meters of a crossing we must use it. But these rules could at times also be ignored in circumstances similar to those for traffic lights. There will, however, be other occasions when it would be worthwhile walking a greater distance than is prescribed in the road code instead of waiting for a break in the traffic flow.

There are several methods available to assist in specifying the logic of systems of which the so-called flow chart is probably the most venerable, its use certainly pre-dates electronic computers. Special variants have been developed for use with expert systems, but in its simplest form the initial part of a chart for our system might look something like that in Figure 6.3a. It will be observed that in both cases the chart leads us back to the start point if the decision is not to cross immediately. This may, at first sight, seem unnecessary. It would have been wrong, however, to have drawn the left hand side of the chart as in Figure 6.3b. It is quite possible that the traffic officer might come to the end of his tour of duty and leave before signalling us to cross. In this case we would be left waiting in an endless loop or at least having a very long wait until the next time a traffic officer came on duty. Equally for one reason or another the control of traffic at our traffic lights may have been taken over by a traffic officer. He would not look favourably on our attempts to cross the street when he signalled us not to do so even if the lights said ‘CROSS’. These finer points are not always readily apparent if we write our specification in narrative form and some sort of logical charting is a great help in bringing considerations such as these to our attention.

A final point worth noting is that the question ‘IS IT SAFE TO CROSS?’ will need quite a lot more definition but we do not do it here. This is because we know that this question will also be used at other decision nodes in the chart. It is a ‘subroutine’ and, as such, is defined seperately. When the program is running it will branch to the sub-routine when required, do the necessary processing, and then return. In this way the sub-routine need only be written once. In our case the sub-routine will contain amongst things the ‘LOOK RIGHT, LOOK LEFT’, etc we identified in our first examination of the problem.

Readers may care to extend the diagram a little further. Those who find this rather too easy may like to have a shot at writing some of the resulting programs. Here those who have access to program languages which specialise in this area, will find the task somewhat less daunting than those who have to rely on BASIC. Incidently PROLOG one of the languages mentioned earlier, will now run on some micro computers.

But, just a minute before we start, have we forgotten something? What about pedestrian overbridges and subways? Back to the drawing board!

Expert systems are thus here to stay. The challenge and the benefits that accrue will dramatically change the way our doctors, solicitors and other professional people go about their work. They will:

Need to commit less of their knowledge to memory.

Need to have fewer reference books on hand.

Have immediately available expert advice that previously had been obtained only after referring the matter to a specialist. Need to spend less time in attempting to keep up-to-date.

The use of expert systems will save professional practitioners a lot of time that currently they must devote to the less than productive activities listed above. I would hope that, as a consequence it would also mean that they could then be able to spend more time on the human relations side of their practice. Could it be that once again general practitioners will get to know their patients when they are well and not merely as someone who has a medical problem? Or that the family solicitor will return to helping his clients deal with the not strictly legal but nevertheless worrying problems that beset all families from time to time?

As an additional bonus, expert systems will open up opportunities for New Zealand specialists, not only doctors and lawyers but also all those in fields where New Zealand plays a leading role, to pass on their skills. The computer industry is preparing to do its part. I hope, now that New Zealand is entering its second quarter century of electronic computing, that the professions will soon be seen to be doing likewise.