Will we let expert systems into our lives?

Written by David Tebbutt, MacUser Dec 1988 (guess) - scanned

I suppose it's inevitable that, one day, you'll sit down at your Macintosh and ask it for some advice. Perhaps your love life will have hit problems or you'll be experiencing a puzzling illness. The Mac will be full of sympathy and, through a series of gentle questions, will arrive at a diagnosis. It may even suggest a cure. A highly developed program may go right ahead and write a 'Dear John' letter for you or print a medical prescription.

Would you feel confident that the program running on the Macintosh was capable of taking things to their logical conclusion? Or would you rather it talked you through your problems, came up with a few suggestions and let you make the final decisions yourself?

What about computers which watch share prices or currency values, trying to capitalise on temporary fluctuations in the market? Automatic trading takes place at the speed of light and, in this situation, buying and selling decisions are made much faster by a computer than a human being. The company with the best program will end up making the most money.

Buying and selling will be done, not on the basis of any underlying reality, but on the basis of tiny natural or computer-induced changes in share prices or currency values. You could imagine a situation where all the computers think they've spotted a downturn and start to sell. With none of them buying the resulting crash would knock October 1987's into a cocked hat.

Satellites gather data and feed it to ground based computers for all sorts of purposes. Our weather forecasts are based on this information but we've seen what happens when weather satellite information is 'interpreted' by humans. The hurricane that occurred here in 1987, was predicted by computers, and not believed. The result was that we were unprepared for the devastation.

Imagine what would happen if a spy satellite thought it saw a bunch of misiles heading towards the USA. The computers munching through the incoming data stream would present their analysis of the situation and by the time it had been discovered that the missiles were actually a flock of birds it could be too late.

Split second decisions are also necessary in the money markets, and in pursuit of an ever finer margin in an ever shorter time, automated computer trading is becoming increasingly common. No doubt their influence will continue to spread until there is no need for maniacs to spend all day scanning computer screens and screaming down telephones. Computers do that more quickly, and definitely more quietly.

These programs are 'expert systems'. They can do almost anything from checking for car engine faults to advising DSS claimants on benefit entitlements. Soon, we will all be affected by expert systems.

In anticipation of this, the Council for Science and Society has published genefits and Risks of Knowledge-Based Systems. It attempts to identify the impact on society of expert systems and makes recommendations to minimise the abuse of these programs.

Among the expert systems it describes is a battle management system and the alarming observation is this: "Because of the complexity of the system and the impossibility of testing it in a realistic setting, we can never know whether a battlefield knowledge based system will work properly until it is too late."

Apparently the Office of Technology Assessment within US Congress has identified three attitudes among artificial intelligence researchers. (Expert systems form part of the discipline.) They are Believers, Social Critics and Disbelievers. The Believers see profound social changes arising from the introduction of intelligent systems. They see improvements in justice, education, work and in empowering the individual. They expect machines to take over menial tasks, and, later, perhaps more complex tasks as well.

The Social Critics believe that intelligent systems may succeed technically but that they will reinforce existing power structures and will have negative social implications.

The Disbelievers think that many of the goals of artificial intelligence are not achievable, not least because human knowledge and behaviour is difficult to formalise.

I think the Believers are fooling themselves when they claim the world will become more democratic. Until quality of life is incorporated into the world's balance sheets, I fear for individuals displaced by semi-intelligent systems.

The Social Critics are probably much nearer the mark. The Disbelievers are probably partially right, in that you cannot reproduce the richness of human understanding in silicon. On the other hand, you can create jolly useful advice-giving programs working within specific knowledge domains.

The bottom line of the book seems to be that, although positive advantages can be found for knowledgebased systems, don't underestimate the huge damage that can be caused by implementing a bad system in a high risk area. It shows five ways in which expert systems can fail.

The authors propose legislation to ensure that no one is misled or in any doubt about the nature of a publicly available expert system. The book also introduces a code of practice for workers in the field.

If you read the book, and it's available from Oxford University Press at £6.95, you will at least be prepared for the day when your Mac becomes your counsellor and confidant.