What’s An Agent, Anyway? A Sociological Case Study

摘要

The term has been picked up, widely appropriated, and in many cases misappropriated, by technical publications, lay publications, and many researchers in computer science. I examine some of appropriations and misappropriations, talk about their reasons and implications, and delve into a case study of an agent that is used to illustratively define what I consider an agent to be, and why many of current uses of term are misleading. The agent I consider is Julia, a TinyMUD robot of Maas-Neotek family, all of whom have been in almost constant use at several sites on Internet for last two or three years. I also speak at length about sociology of such agents, what we might expect to find in future, and why sociology is such an important aspect to consider when investigating programming.1 and appropriation There has been a flurry of references to in both lay and technical press in recent months. While idea of computational agents that do one’s bidding goes back decades,2 explosion of recent press has been astonishing. Pattie Maes and her group’s work on agents3 has been cited in at least two dozen publications in last year or so, many of them lay press (e.g., The Los Angeles Times, Mass High Tech or semi-technical press (e.g., MacWeek, MacWorld, etc). symposium at MIT on agents and their programming, held in October of 1992, drew at least a thousand participants. Further, buzzword frenzy has hit industry as well. We have, for example, a Macintosh program called Magnet, recently released from No Hands Software, which bills itself as the first intelligent agent for In fact, program is essentially a file-finder, which can locate files matching certain characteristics and perform some operation on set, as well as other utilities such as synchronizing a PowerBook filesystem with that of some home, stationary Macintosh. 1This document is Agents Memo 93-01, available from Group, MIT Media Lab. Copyright (C) May 1993 by Leonard N. Foner. Permission to redistribute for academic purposes granted provided that this notice is not removed. Author’s address: [electronic] , [physical] MIT Media Laboratory, E15-305, 20 Ames St, Cambridge, MA 02139. 2Vannevar Bush spoke of such a thing in late fifties and early sixties, as did Doug Englebart. 3A learning interface as opposed to any other kind of is one that uses machine-learning techniques to present a pseudo-intelligent user in its actions. 2 What’s an Agent, Anyway? ________________________________________________________________________________ We also have Your from Bright Star, whose promotional literature starts with, Remember excitement you felt first time you turned on a Mac? Now you can relive magic and realize its full potential with At Your Service, first animated Desk Accessory to give your Mac a ’human’ personality. At Your Service features Phil,4 your Personal Assistant...he talks...he listens...he works for you! In fact, this program is pretty trivial: it utters a spoken greeting when Mac is turned on, issues reminders for preset events, issues alerts when email arrives, reminds you to stop typing every few minutes to ease hand and eye strain, contains a screen saver, and so forth. At Symposium on at MIT in October 1992, Irene Greif of Lotus presented on a groupware version of Lotus 123 that made collaboration between multiple spreadsheet users easier. This is certainly respectable technology, but I would hardly call it an agent, and her presentation had to work fairly hard to force it into mold of an agent-oriented piece of work. The popular imagination about agents that do our bidding has even extended to taking attributing human emotional and intentional states, out of context, to programs and shamelessly running with them. For example, consider Michael Shrage’s article in The Boston Globe, March 7, 1993, imported by Los Angeles Times Syndicate, entitled, A New Dimension in Deception. In this article, Shrage picks up on fact that a calendar-scheduling agent may have to decline a pro