Modeling Human-Agent Interaction with Active Ontologies

As computer systems continue to grow in power and access more networked content and services, we believe there will be an increasing need to provide more user-centric systems that act as intelligent assistants, able to interact naturally with human users and with the information environment. Building such systems is a difficult task that requires expertise in many AI fields, ranging from reasoning, planning, scheduling, natural language and multimodal user interfaces. In contrast to many approaches to building agent assistants where diverse AI components are stitched together at a surface level, we propose an approach, called ”Active Ontologies”, and a toolset, called ”Active”, where a developer can model all aspects of an intelligent assistant: ontology-based knowledge structures, servicebased primitive actions, composite processes and procedures, and natural language and dialog structures. We demonstrate this approach through an example prototype of an intelligent meeting scheduling assistant that communicates using instant messages and emails.