Capability Representations for Brokering: A Survey

In this article we review knowledge representation formalisms that lend themselves to the representation of capabilities of intelligent agents The aim of representing capabilities is of course that we want to reason about them The reasoning task we are most interested in is capability broker ing i e the task of nding an agent which has a capability that can be used to address a given problem Thus the rst area we review here is agent cooperation and communication from which the problem originates However the capability representations found here often lack sophistic ation and are rather ill de ned Thus we have turned to logics next which include some of the best understood representations in AI and bey ond These representations are rather generic though Capabilities are essentially the actions an agent can perform and thus we review action representations next While these representations are very promising they lack the expressiveness required to represent certain more complex capab ilities Models of problem solving address this issue for complex reasoning capabilities and this is the nal area we review in this paper The Problem of Capability Brokering One approach to achieving arti cial intelligence is the rational agent approach Russell and Norvig page In this approach AI is viewed as the study and construction of rational agents An agent is described as an entity that perceives and acts Rationality means that it acts as to achieve its goals given its beliefs More precisely Wooldridge and Jennings page identify four properties an agent should have autonomy social ability reactivity and pro activeness Pro activeness means that an agent should be able to exhibit goal directed behaviour by taking the initiative Pro activeness is directly related to rationality Social ability the property we will be most concerned with means that an agent interacts with other agents possibly humans via some kind of agent communication language Together pro activeness and social ability imply that an agent should communicate not with just any other agent but with those agents that can help it achieve its goals But rst it has to nd these agents This problem is very similar to what Davis and Smith page call the connection problem in distributed problem solving One assumption they are making is that the agents that exist are xed We will assume here that an agent exists in a dynamic environment with other agents As the environment changes new agents might come into existence or existing agents might disappear Agent autonomy means that an agent has to operate without the direct intervention of humans i e that it has to nd out by itself about existing other agents speci c ally agents that can help it achieve its goals Genesereth and Ketchpel page distinguish two basic approaches to the connection problem direct com munication in which agents handle their own coordination and assisted coordin ation in which agents rely on special system programs to achieve coordination The best known work in AI on agent communication is probably the Knowledge Sharing E ort Fikes et al Neches et al Part of this e ort is the development of the Knowledge Query and Manipulation Lan guage kqml a high level agent communication language Finin et al Finin et al Labrou and Finin kqml like most approaches to the connection problem advocates assisted coordination through facilitators and me diators However the support o ered by kqml for this task is still an active research issue especially for more complex agents Kuokka and Harada b Decker et al have recently described a solution space to the connection problem that identi es nine di erent types of middle agents depending on which agents know about capabilities and preferences of agents initially By a prefer ence they mean meta knowledge about what types of information have utility for a requester In a solution to the connection problem in which capabilities are ini tially known to the provider and the middle agent only and in which preferences are initially known to the requester and the middle agent only the middle agent is what they call a broker Capability brokering and more speci cally represent ations for capabilities that facilitate brokering are what we shall review in this article