Message-Id: <199308311257.AA02902@eli.CS.YALE.EDU>
Date:         Tue, 31 Aug 1993 08:54:02 EDT
Reply-To: "Robert J. Chassell" <bob@GRACKLE.STOCKBRIDGE.MA.US>
Sender: Lojban list <LOJBAN%CUVMB.bitnet@YaleVM.YCC.YALE.EDU>
From: "Robert J. Chassell" <bob@GRACKLE.STOCKBRIDGE.MA.US>
Subject:      Believable Agents
To: Erik Rauch <rauch>
Status: O

The recent discussion on attitudinals leads me to repost the following
discussion regarding "Believable Agents".

To what degree does the expression of attitudinals in Lojban provide
the listener with a sense that the speaker is ``"really there" --
aware, intentioned, and capable of significant social interaction.''?

In English, attitudinals are often expressed through body language and
tones of voice.  In Lojban one can express attitudinals using words
only.  Does this format make it easier to consider some of the
questions raised in the discussion below?

    Robert J. Chassell               bob@gnu.ai.mit.edu
    Rattlesnake Mountain Road        bob@grackle.stockbridge.ma.us
    Stockbridge, MA 01262-0693 USA   (413) 298-4725


AAAI 1994 Spring Symposium Series
Stanford University, California
(Special Symposium, March 19-20)

Believable Agents

AI has long sought to construct autonomous creatures. The thought
of these entities brings special delight when they are imagined to
project a sense of being "really there" -- aware, intentioned, and
capable of significant social interaction. Through an analogy
presented below, let us call these "believable agents."

Many research groups are now trying to build "complete" agents.
Much of this work occurs under the themes of situated agents and
integrated architectures, but relevant work occurs across varied
subareas such as intention, emotion, and discourse.

As computational technology for "complete" agents has developed,
demand for believable agents has increased in such domains as
user interfaces and interactive entertainment. Meeting this
apparently difficult demand might not require accurate modeling
of human cognitive processes. It might not be necessary to build
especially intelligent or competent agents, measured in some
objective sense. The actual requirement is to achieve a persistent
appearance of awareness, intention, and social interaction.

The arts (film, theater, novels, radio, drama) have long studied a
related problem. Here the agents are called "characters," and the
requirement that the characters project the image of being really
there is called "permitting suspension of disbelief." Artists, (such as
traditional animators), know much about how to achieve this using
highly simplified agents (simple body, simple mind, simple
environment, e.g., Elmer Fudd). They know what abstractions can
be made safely, and which facets of appearance and behavior are
crucial to maintain. We believe these artists can convey significant
useful knowledge to AI researchers trying to construct believable
interactive creatures.

The goal of the proposed symposium is to bring together
researchers interested in exploring the idea of "believability" from
an AI point of view, including a selected set of artists who can shed
light on traditional notions of, and techniques for, establishing
suspension of disbelief. We feel this is a fresh way of looking at the
task of building agents, and that clarifying the nature of this task
will yield interesting new research problems and useful new
artifacts.

Among the questions we hope to discuss are: What makes
characters believable? What makes agents believable? Are agents
different from characters? What is the nature of current integrated
architectures and situated agents? How well do they support
believability? Are there clear areas demanding study? How much
breadth of capability is necessary to produce a believable agent?
How much depth (competence) is necessary? What are the
respective roles of movement and language in achieving
believability? What is the role of context in establishing
expectations in the user and thus in simplifying the task? For
instance, how can the setting of an interface agent or the theme of a
story help limit the technical requirements on agents? How can we
measure believability and progress toward believability? Why
don't artists use "scientifically valid" techniques to evaluate the
believability of their characters?

Prospective participants should submit a short summary of their
relevant research and/or artistic activities (two-three pages) that
emphasizes the fundamental goals of the work, the status of the
efforts, and especially its relevance to the construction of believable
agents. Be sure to include a mailing address, telephone number, fax
number, and email address.

The organizing committee will consider the submissions and invite
selected participants to present material at the symposium.


Electronic submissions are preferred, but these must be plain,
unformatted text. If this is impossible, then send hard copy. In
either case, please send one copy to:
joseph.bates@cs.cmu.edu or
Joseph Bates
School of Computer Science
Carnegie Mellon University
5000 Forbes Avenue
Pittsburgh, PA 15213