Communication is a Coordination Problem
Les Gasser
University of Illinois
Historically, Multi-Agent Systems research has
treated coordination as its central issue. Communication is seen
as a critical capability that allows agents to coordinate their
behaviors. Against this frame, MAS communication research has
centered around two principal issues:
- Establishing general semantic theories that
attempt to ensure that communications have predictable effects
if all agents follow the rules (for example as seen in the development
of KQML, FIPA, and underlying semantic models, such as the Cohen-Levesque
theory of joint intentions), and
- Designing standard communication languages,
paradigms, protocols, and toolkits, often based on those semantic
theories, so that theory-based interactions can be embedded
into multi-agent coordination frameworks.
The fact that the ability to communicate itself
is a coordination problem has escaped much attention in the MAS
world. If autonomous agents in open systems need common---or at
least inter-operable---terminologies, linguistic structures (e.g.,
grammars) and semantics as a basis for acting together, how can
the agents themselves derive them and what are the limits? At
present, the coordination needed for communication is done by
people who design agent ontologies, languages, and protocols.
This fundamentally limits agents' autonomy, and forms a kind of
invisible centralization. Our aim is to shift at least part of
this burden to agents themselves, by studying these issues as
problems of large-scale distributed representation, coordination,
and continuous collective design.
[
]
|