Track:
Contents
Downloads:
Abstract:
The purpose of this paper is to extend verification tests to systems with multiple autonomous agent knowledge bases. Using a classic approach to verification, this paper focuses on tests concerned with consistency, completeness and correctness. In particular, this paper focuses on those unique issues that are generated as we go from single agent systems to multiple agent systems. This paper is concerned with inter agent verification, since previous results can be used for intra agent verification. For example, consider one agent with the rule "ifA then B" and a rule in another agent "irA then C". In such a setting, the agents would be constantly at odds. Alternatively we might find the following rules in one agent ("if A then B" and "if C then A"), while another agent that interacts with that agent might have the rule ("if B then C"). With those two interacting rule bases a dialogue starting with "A" could cycle indefinitely. One potential approach to multiple agent systems is to compare the knowledge base of each subset of agents to determine the existence verification issues. Where the number of agents is small this approach is feasible. However, for even medium size systems this approach explodes computationally. This paper finds that many of the multiple agent verification tests can be conducted on a meta rule set generated from all the rules contained in each of the agents’ knowledge bases thus minimizing computational effort. In addition, this paper finds that the property of agent "isolation" is an important verification criteria in multiple agent systems.