Dynamic inference of perceptual categories in negotiating agents

In this paper, we describe an approach by which one agent may infer the categorical preference structure of another agent in relation to a set of alternative proposals made in negotiation. The categorical preferences are inferred over a sequence of proposals, as preferences are disclosed one at a time. We seek to avoid strong ontological commitment between the agents, making only minimal assumptions about shared representations of the set of alternatives. We adopt a multidimensional representation that avoids the common, but rarely plausible, assumption of issue independence. An illustration of the approach is included and possible applications are discussed.