How Can Agents Know What to Assume When?

The work in this paper is motivated from the need for assumption-based reasoning in normative systems, where, realistically, agents will have incomplete knowledge about their environment, and about other agents. The question we seek to address is whether it is possible for agents to identify appropriate assumptions dynamically, in order to fill in informational gaps. We discuss and illustrate our proposal with reference to an e-commerce example. In our previous work, we argued that e-contracts could be represented as Default Theories and proposed a theoretical way in which such theories could be constructed automatically from initial Event Calculus representations. That proposal relied on determining what information could be proved from the agentpsilas knowledge base, in order to decide whether it would serve as an assumption or not. In this paper we present an incremental technique that can be used for this construction that enables the dynamic and ad hoc identification of candidate assumptions, without resorting to proof.