Interface agents are computer programs that employ Artificial
Intelligence techniques in order to provide assistance to a user
dealing with some computer-based task(s). In this talk I will
present several implemented examples of interface agents which
have been modelled after the metaphor of a personal apprentice.
These agents learn how to assist the user by (i) observing the
user's actions and imitating them, (ii) receiving user feedback
when they take wrong actions, (iii) being trained by the user
on the basis of hypothetical examples and (iV) asking other agents
(working for other users) which have more experience for assistance.
I will discuss how these learning agents can be implemented using
memory-based learning and statistical learning techniques. I
will present results from three prototype agents built using
these techniques:
* an agent which assists a user with the scheduling of meetings
(accept/reject, schedule, reschedule, negotiate meeting times,
etc),
* an agent which assists the user with electronic mail (sorting,
prioretizing, forwarding, archiving), and,
* an agent which assists the user with filtering of electronic
news (in particular Usenet Netnews).
The talk will be concluded with some plans for current and
future research.
Pattie Maes received her PhD in Computer Science from the
University of Brussels in Belgium in 1987. In 1989 she came to
the MIT Artificial Intelligence Laboratory to work with Marvin
Minsky and Rodney Brooks on modeling and building Intelligent
Autonomous Agents. In 1991 she joined the MIT Media Laboratory
as an tenure track Assistant Professor. Once at the Media Laboratory,
she started thinking about how her work on Intelligent Agents
could be applied to the problem of Human-Computer Interaction.
In particular, her own frustration with hours of waisted time
reading electronic mail, browsing through electronic bulletin
boards and electronic news and scheduling meetings motivated
her to build computerized personal assistants that help a user
with daily tasks and activities.