AgentPrIMe: Adapting MAS designs to build confidence

The products of systems cannot always be judged at face value: the process by which they were obtained is also important. For instance, the rigour of a scientific experiment, the ethics with which an item was manufactured and the use of services with particular licensing all affect how the results of those processes are valued. However, in systems of autonomous agents, and particularly those with multiple independent contributory organisations, the ability of agents to choose how their goals or responsibilities are achieved can hide such process qualities from users. The issue of ensuring that users are able to check these process qualities is a software engineering one: the developer must decide to ensure that adequate data is recorded regarding processes and safeguards implemented to ensure accuracy. In this paper, we describe AgentPrIMe, an adjunct to existing agent-oriented methodologies that allows system designs to be adapted to give users confidence in the results they produce. It does this by adaptations to the design for documentation, corroboration, independent storage and accountability.