Testing a multi-agent model for alternative plan generation

In the academic field of planning it is commonly agreed upon that there are two elemental needs that planning support systems (PSS) ought to serve: first, an informed process to generate alternative plans, and second, extensive tools for tracing out the consequences of alternatives in order to evaluate them. The abundance of complexity, uncertainty, and subjectivity involved in meeting either of these requirements raises the need for intelligent PSS. Multi-Agent (MA) technology, as one of the branches of Artificial Intelligence, is a powerful instrument that can increase the complex problem-solving capabilities of a system by means of its distributed problem-solving approach. As such, it is pre-eminently suitable for meeting both requirements. Promising advances have already been made in utilizing the technology to serve the elemental planning need of evaluation through forecasting. To date, however, the generation of alternative plans remains relatively underexposed in this perspective. The proposed paper will closely examine the operation of a MA-based tool that has been developed to support planners in easily generating sets of alternative plans for exploratory purposes and studying the effects of domain-specific objectives. The model underlying the tool identifies a group of agents who represent the various stakeholders active in the plan development process and interact in order to cooperatively develop alternatives, starting off from individual viewpoints. In the paper various model properties will be addressed and judged by means of a small series of test cases.