Abstract:
An adaptive agent interaction strategy is
presented. It has been trailed in a system for
emergent process management. That system was
constructed using generic process agents. Their
architecture is a three-layered BDI hybrid
architecture.
The agent interaction strategy substantially
determines the perforniance of a multi-agent
system. The interaction strategy described here
is derived by observing the way that humans
interact in process management applications.
Individuals are driven principally by their goals
but this is tempered by social interaction
attributes such as whether the individuals they
are dealing with are “cooperative ”, “friendly ”,
“trustworthy”, and so on. The strategy
described takes into account sentiments such as
a friendship and trust by making estimates of
them for each other agent in the system. Using
this strategy each agent continually adapts its
behavior on the basis of its observations of other
agents and a continual revision of these
estimates.
The experiment using this approach led to
high performance of the process management
system in the context of the overall operation of
the organization. That is, all using the system
agreed that they fell “comfortable” with way in
which the system operated and that its overall
performance was higher than could be expected
from a manual system.
