Abstract:
In a multiagent system agents negotiate with one another to distribute
the work in an attempt to balance the incompatible goals of optimising the quality
of the result, optimising system performance, maximising payoff, providing
opportunities for poor performers to improve and balancing workload. This distribution
of work is achievedby the delegationof responsibility for sub-processes
by one agent to another.This leads to estimates of the probability that one agent
is a better choice than another.The probability of delegating responsibility to an
agent is then expressed as a function of these probability estimates. This apparently
convoluted probabilistic method is easy to compute and gives good results
in process management applications even when successive payoff measurements
are unpredictably varied.
