Agent Provocateur

Appeared in Volume 9/2, May 1996

Software agents seems to be cropping up in a multitude of guises: they're lurking in operating systems, software engineering, computer vision and LP (I suppose guises come naturally to agents, especially undercover ones).

At the heart of this popularity is the computer scientist's wish to wrench from us everything which makes life novel. The agents of the future will decide what clothes we wear, what we eat, what newspapers we read, what vacation spots we frequent, and so on. Faced with this massive removal of responsibility, I wonder what I'll be left with to do? Perhaps I'll just stay in bed (if my 'bedroom' agent allows me). Agents seem to be designed to return us to the level of 5 year old's, or into the role of Bertie Wooster, ordered around (politely) by a software Jeeves.

Actually, I doubt whether agents will become quiet so domineering. More likely, they'll aspire to the standards of a character familiar to us from 1970's Disaster Movies - the guy who 'gets it' first. He's the one who dies abruptly when he decides it's okay to use the lift to escape, or opens the reactor room door just as the hero shouts 'Stop!'.

Agents of this ilk are fine under normal conditions, but have a habit of making rather poor decisions under unusual, unsettled, circumstances. The drawback is that these decisions affect us: we may be instructed to wear a diving suit today, roller-skate to the office, and spend two weeks holidaying in the broom cupboard.

An unnerving thing about agents will be their number. It won't be you against a single unpredictable agent, but a highly organised, coordinated army, all with your best interests in mind, and the collective means to enforce them. There will come a day when you, and all your colleagues, will turn up to work in colour-coordinated diving suits and roller-skates.

Perhaps I'm being just a tad alarmist. I remember the dour proclamations of disaster associated with expert systems in the early 1980's. These had the required effect of releasing satisfyingly large amounts of research funding into the area, but the actual apocalypse did not occur. Particular expert systems have been successful, but the grandiose, over-arching expert systems, scheduled to control every aspect of our waking hours did not (thankfully) materialise.

It is reassuring to compare the claims made about agents with those for expert systems: they are very similar; we can also hope that the outcomes will be the same.

Andrew Davison