Distributed production systems

Abstract

Techniques for building problem solving systems capable of adapting to changing environmental conditions are of great interest. This chapter has presented an approach that relies on reorganization of a collection of problem-solvers to track changes in deadlines and problem solving requests. The approach exploits an adaptive trade-off of resources and organization form to satisfy time and performance constraints. Agents are created and destroyed, and domain knowledge is continually reallocated. To extend the possible architectures for organization self-design, composition and decomposition have been introduced as new reorganization primitives. Organizational knowledge has been formalized to represent interactions among agents and their organization.

With additional decision-making meta-knowledge, this approach can become a more general organization self-design technique. It also has the advantage of being grounded in a well-understood body of theory and practice: parallel production systems. In the current version, composition/decomposition decisions are made solely on the basis of firing ratios, and the choice of rules to transfer is made arbitrarily. Note, however, that allocation decisions could instead be based on the semantics of rules (i.e., distribution based on the kinds of tasks that need more resources). Moreover, partial knowledge transfer among existing agents can be combined with the composition and decomposition to provide a flexible and distributed task-sharing system.