Distributed Artificial Intelligence

of the choice and could be used to plan which observations to perform next to reduce the

uncertainty.

Our initial design had only one hypothesis contained in each Activation Framework

Object and the system used the activation level of that hypothesis as the criterion for

which AFO to execute next. It quickly became obvious that such a structure would result

in too much fragmentation with too much message traffic. As a result multiple hypotheses

are now embodied within an AFO. Each hypothesis has a name, a data structure, and an

activation level which specifies the belief in its truth (positive activation levels) or falseness

(negative activatio ...