Reacting to unexpected events and communicating in spite of mixed ontologies

Abstract

We describe our experiences in building agents (and their environment) that could solve important problems in agent to agent communication, in manners that are not pre-programmed or reactive: (1) Two agents may have different ontologies. We do not assume that agents share the same classification of concepts: they have each one its own ontology or concept categorization. Agents can not exchange concepts: they have to exchange symbols (words of natural language), which the receiving agent has to map to the intended concept, for interaction to be meaningful. (2) The model for interaction. Our agents interact through scripts or frames having roles, resources and prerequisites. Agents select which roles in what scripts to play, in order to satisfy their purposes. (3) Unexpected events happen and throw the agents out of their current plan or execution. (4) Planning is needed in this changing world.