A Quantitative Model of Capabilities in Multi-Agent Systems

Reasoning about capabilities in multi-agent systems is crucial for many applications. There are two aspects of reasoning about the capabilities of an agent to achieve its goals. One is the symbolic, logical reasoning, which is based on whether the agent can determine a plan that can be used to achieve a goal (i.e. “know-how”). Another is quantitative reasoning about levels of skill and whether a set of tasks can be achieved within a specified time and with quality constraints. Capabilities in this sense are determined by the limits of internal processing capacity. Both artificial agents and human agents can be subject to processing capacity limits, whether due to CPU speed, or cognitive limits on attention, memory, etc. This could depend not only on the unique skill levels of each individual and their cognitive abilities, but also on their ability to handle those task demands in parallel, given intrinsic limits on internal cognitive resource capacities. In past work, researchers have focused primarily on the logical aspect of capability reasoning; less work has been done on modeling the quantitative aspects of reasoning about capabilities in agents. In this paper, we introduce a general mathematical model to define the capabilities of agents to achieve a set of tasks. Our definitions of capabilities are based on whether a feasible schedule exists to complete the tasks within the constraints, either in static environments or dynamic environments, for which we present two corresponding preliminary scheduling algorithms. We illustrate this model with two experiments to evaluate the algorithms. We conclude by discussing the potential applications of this model, and future work.