Loren Terveen (1), Markus Stolze (2) and Will Hill (3)
(1)
AT&T; Bell Laboratories
600 Mountain Avenue
Murray Hill, NJ 07974
terveen@research.att.com
(2)
The University of Colorado
Computer Science Dept.
Box 430 Boulder, CO 80309
stolze@cs.colorado.edu
(3)
Bell Communications Research
445
South Street
Morristown, NJ 07962
willhill@bellcore.com
Direct manipulation is the outstanding success story and dominant paradigm in modern interface design. A model world is created, in which graphical objects represent objects in the domain of interest, and users perform domain actions with gestures of a pointing device. Intelligent agents usually are seen as a competitive paradigm, most likely because agents typically are presumed to communicate using natural language.
A very common and important role for interface agents is to assist
users in performing design tasks. While a user creates a design using a
direct manipulation interface, an agent can compute assistance such as
detecting problems with the design, suggesting alternative or related
issues to consider, or automatically carrying out routine or low-level
aspects of the design task. Such information usually has been
communicated in natural language. However, this has a number of
disadvantages:
(1) generating good text is a difficult computational problem,
(2) interpreting text is a non-trivial cognitive task for users,
(3) textual methods usually present information sequentially and may
be interruptive and intrusive, and
(4) text is a separate medium, so the existing interaction context --
the graphical objects -- is not exploited.
We are interested in exploring methods that enable agents to deliver their assistance by manipulating the display of graphical objects. Agents need not be "talking heads"; rather, they can communicate much information by adding or deleting graphical objects, enhancing the display of existing objects by coloring or shading them or changing fonts or other visual properties, changing cursor shape, and graphically depicting relationships between objects. These communication techniques exploit human perceptual abilities and the unique communicative potential of computational media, e.g., to deliver information in parallel and without interrupting users, and reduce the problems of text generation and interpretation. They take advantage of the fact that human collaboration often is centered around artifacts. Finally, centering communication around graphical objects has the potential to reduce the dangers of users anthropomorphizing computational agents, attributing human-like abilities to them.
In short, we seek to turn model worlds into magic worlds, which change not only in response to user actions, but also due to the workings of hidden, helpful hands. Design objects seem to take on a life of their own, as agents working on behalf of the user's goals add, delete, and modify objects as approriate. Thus, usefulness to the current user task -- not only accurate simulation of (an aspect of) the real world -- becomes the primary criterion for determining the appearance of the interface. The primary goal of the workshop is to explore and clarify the nature, scope, limits, and dangers of the ``magic world'' paradigm. We consider a number of specific issues, such as: