Practical Reasoning About Complex Activities

In this paper, we present an argument-based mechanism to generate hypotheses about belief-desire-intentions on dynamic and complex activities of a software agent. We propose to use a composed structure called activity as unit for agent deliberation analysis, maintaining actions, goals and observations of the world always situated into a context. Activity transformation produces changes in the knowledge base activity structure as well in the agent’s mental states. For example, in car driving as a changing activity, experienced and novice drivers have a different mental attitudes defining distinct deliberation processes with the same observations of the world. Using a framework for understanding activities in social sciences, we endow a software agent with the ability of deliberate, drawing conclusion about current and past events dealing with activity transformations. An argument-based deliberation is proposed which progressively reason about activity segments in a bottom-up manner. Activities are captured as extended logic programs and hypotheses are built using an answer-set programming approach. We present algorithms and an early-stage implementation of our argument-based deliberation process.