Architectures and Idioms: Making Progress in Agent Design
Date Friday, 29JUN00
Time 2-3pm
Speaker Joanna Bryson
Affiliation MIT AI Lab
Abstract This paper addresses the problem of producing and maintaining progress in the discipline of agent design. The proliferation of architectures and methodologies is valuable for the ideas and experience brought to the field. Unfortunately, these ideas can be difficult to harness because, on real projects, switching between architectures and languages carries high cost. In this paper we propose that this cost can be reduced if the research community takes the responsibility to reexpress their innovations as idioms or extensions of one or more of the current de facto standards. We describe as an example the concept of a Basic Reactive Plan, an idiom that recurs in many influential agent architectures, yet in others is difficult to express. This proposal and example shed light on the relationship between the roles of architectures, methodologies and toolkits in the design of agents.

2nd half: informal overview of evaluating and comparing architectures, including: 1) proof of concept in blocks-world and on a mobile robot, 2) statistical hypothesis testing in a simulated environment, and 3) analyzing trends in the complex agent literature.
Location 545 Technology Square (aka "NE43")
Room 8th floor play room