Transporting the Agent vs. Transporting the Interface

			   Henry Lieberman
			    MIT Media Lab

The most prevalent notion of a "transportable" agent is one that
consists of code initially running on one machine being downloaded and
continuing to run on another's machine. But from the user's
perspective, what difference does it make [except perhaps for speed]
where the code actually runs, as long as it accomplishes the task the
user asks of it? From this point of view, transportability is simply
an optimization technique and not a fundamental characteristic of
agent software.

However, interfaces for "surfing" the Web present a different
perspective on the transportability issue. Users DO get a strong
feeling of motion or travel while browsing the Web, as witnessed by
the plethora of spatial metaphors: "visiting" a Web site, being "at" a
Web site, a Web site having "moved", etc. What accounts for this
difference? To have truly "transportable" agents, the user should
be able perceive the "motion" of the agent in a manner like that in
which he or she perceives the motion of surfing through the Web. 
We illustrate this with a description of Letizia, an agent for
assisting the user in browsing the Web, that acts as an advance scout,
staying just a little bit ahead of the user's browsing, learning from
the user's behavior, and making real-time recommendations of new pages
of interest.