Negotiating Socially Optimal Allocations of Resources with Argumentation

The resource allocation problem of multi-agent systems is the problem of deciding how to allocate resources, controlled by agents, to agents within a given system. Agents typically have preferences over alternative allocations of resources. These preferences may be derived from the agents’ goals, which can be fulfilled by different plans (sets of resources). The problem arises because agents may not be able to fulfil their goals without being re-allocated resources controlled by other agents and agents may have conflicting preferences over allocations. Examples of the resource allocation problem include electronic commerce (where resources are commodities equipped with prices), the grid (where resources are computational entities equipped with computational power), and scheduling and timetabling (where resources may be tasks with costs). The focus in this thesis is distributed decision-making amongst agents, whereby agents actively participate in computing re-allocations, starting from initial allocations which may or may not fulfil their goals. A re-allocation is arrived at by means of local negotiation steps wherein resources change hands between the agents involved in the negotiations. The negotiation method of choice in this thesis is argumentation-based negotiation supported by assumption-based argumentation. This method allows agents to work towards their goals despite incomplete information regarding the goals of and resources allocated to other agents, to share knowledge, thereby eliminating unknowns, and to resolve conflicts within themselves and between one another which may arise because of inconsistent information. Solutions generated by a resource allocation mechanism may be ranked according to how they affect the individual welfare of the agents as well as the overall social welfare of the agent society, according to different notions of social welfare borrowed from economics. The argumentation-based negotiation mechanism we propose guarantees, for the problem domain of interest in this thesis, that negotiations between agents always terminate converging to a solution. Moreover, the mechanism guarantees that solutions reached optimise the welfare of the individual agents as well as the agent society as a whole according to Pareto optimal and utilitarian notions of social welfare