Expert Systems ’85 (conference review)

Transcribed from the printed issue; not yet reviewed. Read it in the PDF of the issue, page 77

“Expert Systems 85” was the somewhat unoriginal name given to the 5th Technical Conference of the BCS Expert Systems Specialist Group. Frankly, it wasn’t much different from any of the previous conferences except that it was bigger – over 900 eager beavers attended.

Assembled gurus included all the old familiar figures, though – Alex d’Agapayeff late of CAP and now of Expertech, Martin Merry late of GEC and now of Hewlett-Packard, Aaron Sloman of Sussex, Father Aloysius Hacker of Hacker Enterprises, Austin Tate of almost everywhere, and so on down the list of experts about expert systems.

As well as the 900 conference attenders, many of the day trippers who set off to visit the accompanying exhibition, also at the University of Warwick, actually arrived. (Coventry is 10 miles from Warwick, but “University of Coventry” doesn’t sound as good as “University of Warwick”). The throng having been thus swelled by survivors of the cross-country trip, the exhibition was pretty sticky for much of the time, and difficult to gather information (though I did witness the theft of a complete knowledge base by a daring young man. He had enquired about the quality of documentation, settled down to examine a proferred listing and then, when the salesman’s back was turned, coolly walked off with it!). Portability of extremely valuable knowledge is one of the hazards of expert systems.

Everything was for sale from a £20 demonstration expert system, through videotapes, software, consultancy and robots to hardware systems costing £250,000 or so (every desk should have one??)

The conference proper followed the modern fashion for two or more parallel sessions. As usual this resulted in the simultaneous coverage of interesting topics in two different places, followed by the simultaneous availability of two equally depressing longueurs. Each paper in a session was supposed to finish at the same time as the corresponding one in its parallel ‘stream’ but they always over-ran or under-ran by several minutes, so that the closing stages of one, or the essential explanatory moments of another, lecture were constantly interrupted by multiple noisy, out-of-breath arrivals and departures. Corridors and stairs between the theatres would be filled by surging crowds, moving in opposite directions, having missed the dénouement of one lecture in their haste to be able to understand what the next man was talking about. And some unfortunates haunted the passageways, desperately shuttling between alternate attractions like the donkey between two piles of straw, unable to decide which to enjoy.

Enjoyment is perhaps the wrong word to use of papers such as “Symbolic uncertain inference: a study of possible modalities”, though a certain innocent amusement arose from time to time – as when one chairman, by giving an eloquent introduction to a talk on the application of casual knowledge to real-time process control, revealed that he had failed to do the most basic homework of reading the paper (which was actually about causal knowledge!). This latter paper, castigated by some as indecently non-academic, actually showed a serious application of deeper, causal, knowledge as being necessary to the proper handling of information overload situations such as North Sea oilfields (or, in retrospect, the Three Mile Island disaster, where 500 warning lights went off or on in the first minute alone!).

Whilst many papers should never have been let out of their ivory towers, the most pragmatically useful were not, as might have been expected, on techniques and applications, but tended to be about the user interface – for example users quite often don’t just need a diagnosis but need remedies (or even explanations as to why what the user did, as a remedy, actually worked). In an extremely entertaining lecture on an unusually readable paper, a team from Bath reported on expert systems as a medium for knowledge transfer (rather like d’Agapeyeff’s “sharing know-how”) and how human knowledge is a lot more than just the information and heuristics they are supposed to be by traditional expert systems and knowledge elicitation processes. They showed convincingly that humans use tacit knowledge without ever realising it, and that skilled accomplishments may need analysing in order to make sense of the formal knowledge currently being collected.

One of the examples of tacit knowledge they used was this step from a recipe: “Beat the egg whites until stiff and then fold in” – just think about explaining that to a computer. . . . which bit of an egg is white? (not the egg white, which is transparent!), which bit is an egg white? how do you get it out? what does “beat” mean, or “stiff” or “fold in”? Although this is an extreme example of human-speech being difficult to formalise, I’m currently finding out the hard way that there’s an awful lot of tacit knowledge about everyday information sources which it’s difficult to disentangle and formalise, and so am realising that Expert Systems ’85 wasn’t quite as academic as it seemed at the time. Overall rating for people with a slight interest in expert systems? . . . . worth a detour – perhaps not worth a special trip.