Coglunch - October 13, 2005
"Whither Autonomous Agents?"
Carl Hewitt
MIT
Autonomous agents have an interesting history. They first became
technically feasible with the genesis of the field of Artificial
Intelligence in which most of the founding researchers were very
optimistic about prospects for success. Then computers were connected
with networks so the field of Distributed Artificial Intelligence was
born and many researchers were still optimistic. Various approaches
were explored with some initial success. But fundamental problems
were discovered and progress slowed. Distributed Artificial
Intelligence evolved into distinct subfields Multi-Agent Systems and
Autonomous Agents, each with its own emphasis and its own conference.
After a high US government research sponsor complained about the
multiplicity of conferences, the two subfields used this as a
"convenient excuse" to join in a single conference AAMAS.
This talk analyzes the following fundamental dichotomies involving
autonomous agents, and explores their tradeoffs:
- Actors versus agents
- Interdependence versus autonomy
- Commitments versus beliefs, desires, and intentions
- Concurrency versus sequential steps
- Indeterminacy versus choice
- Physics versus mathematical logic
- Negotiation versus reasoning
- Process versus rationality
- Meaning versus behavior
- Truth versus method and practice
- Trust versus circumspection
- Psychology versus sociology, anthropology, and the philosophy of science
Last modified: Fri Feb 10 11:17:44 PST 2006 by emma@csli.stanford.edu