James L. Crowley (Grenoble INP)

In their debate at ACM CHI 97, Ben Schneiderman and Patti Maes opposed intelligent agents to direct manipulation as the "desirable" dominant paradigms for human-computer interaction. Maes argued that intelligent agents were necessary to compensate for the growing complexity of pervasive digital devises, citing the infamous problem of ubiquitous VCR clocks blinking 12:00 because setting the time was too complex. Ben Schneiderman countered that early attempts at intelligence had proved to be counterproductive and in fact disruptive, and that it would be more effective to empower users with new tools for direct interaction with systems.

While this debate occurred more than 15 years ago, the questions have gained in pertinence as technology has matured. In this talk I argue that direct manipulation and intelligent agents are are two distinct members of a larger taxonomy of interactive services, and that each member has its usefulness and domains of application. I present this larger taxonomy, and then concentrate on the class of socially intelligent interactive agents.

If we examine Turing's definition of intelligence, we see that for Turing, intelligence is a description that people use to describe the interaction of an agent with its environment. Rod Brooks has argued that to be considered as intelligent as system must have three properties: It must be embodied, it must be autonomous and its behaviour must be situated. I argue that to be effective as a partner, an intelligent agent must exhibit socially situated behaviour. Current research on intelligent interaction has focused on recognition and communication of linguistic signals while most human-to-human social interaction is non-verbal and highly dependent on social context. A technology for intelligent interaction must be able to perceive and assimilate non-verbal social signals, to understand and predict social situations, and their consequences and to acquire and develop social interaction skills. I conclude by speculating that the planetary scale data-base of human social interaction offered by digitized media (including the internet), coupled with distributed learning from communication between large numbers of interactive agents may offer two new enabling technologies that can provide models for social interaction.

A. M. Turing. "Computing machinery and intelligence". Mind 59.236: pp 433-460, 1950.

B. Shneiderman, Ben, and P Maes. "Direct manipulation vs. interface agents". Interactions 4.6: pp42-61, 1997.