Transcript Document
Gheorghe Tecuci1,2, Mihai Boicu1, Dorin Marcu1 1 Learning Agents Laboratory, George Mason University 2 Center for Strategic Leadership, US Army War College IJCAI-03 Workshop on Mixed-Initiative Intelligent Systems Acapulco, Mexico, 9 August 2003 Disciple: an approach to KB and agent development Develop a learning agent shell that can be taught directly by a subject matter expert to become a knowledge-based assistant Use several levels of synergism between the expert that has the knowledge to be formalized and the agent that knows how to formalize it, and between complementary learning methods: The expert teaches the agent The agent learns from the 1. Mixed-initiative problem solving 2. Teaching and learning 3. Multistrategy learning to perform various tasks in a way that resembles how the expert would teach a person. expert, building, verifying and improving its knowledge base Main idea of the Disciple mixed-initiative approach The complex knowledge engineering activities, traditionally performed by a knowledge engineer (KE) and a subject matter expert (SME), are replaced with equivalent activities performed by the SME and a Disciple Agent, through mixedinitiative reasoning, and with very limited assistance from the KE. Define domain model Create ontology KE SME Define rules Traditionally Define initial model KE SME Extend domain model SME Agent With Disciple Import and create initial ontology KE SME Agent Specify instances SME Verify and update rules Learn ontological elements Agent Define and explain examples SME Agent Learn rules Agent Critique examples SME Explain critiques SME Agent Refine rules Agent Current status: Parallel KB development experiment – Sp03 Initial KB Domain analysis and ontology development (KE+SME) Parallel KB development (SME assisted by KE) DISCIPLE-COG Team 1 stay informed be irreplaceable 5 features 10 tasks 10 rules Knowledge Engineer (KE) All subject matter experts (SME) 37 acquired concepts and Extended KB features for COG testing DISCIPLE-COG Team 2 communicate 14 tasks 14 rules DISCIPLE-COG Team 3 be influential 2 features 19 tasks 19 rules KB merging (KE) Unified two features Deleted 4 highly incomplete rules Refined 11 rules Did not affect the other 84 rules +9 features 478 concepts and features +105 tasks 134 tasks +95 rules 113 rules Correctness = 98.15% 432 concepts and features, 29 tasks, 18 rules For COG identification for leaders Training scenarios: Iraq 2003 Arab-Israeli 1973 War on Terror 2003 DISCIPLE-COG Team 4 have support 35 tasks 33 rules DISCIPLE-COG Team 5 be protected be driving force 3 features 24 tasks 23 rules Learned features, tasks, rules Integrated KB DISCIPLE-COG 2.5 examples/rule 5.47 hours average training time COG identification and testing (leaders) Testing scenario: North Korea 2003 Envisioned life-cycle of future Disciple-MI Building an agent shell DISCIPLE-MI Knowledge engineer 1 Agent doctrinal training Knowledge base optimization and re-use DISCIPLE-MI 2 6 DISCIPLE-MI Domain experts Knowledge engineer Intelligent tutoring 3 Agent Lifecycle DISCIPLE-MI 5 After action review and agent personalization DISCIPLE-MI 4 Agent use and non-disruptive learning DISCIPLE-MI Research directions Modeling expert’s reasoning Learnable knowledge representation Multistrategy teaching and learning Acquisition of expert’s language Mixed-initiative problem solving Resource-bounded learning Learning a model of the expert User-agent interaction KB optimization and integration Intelligent tutoring Acknowledgements This research was sponsored by the Defense Advanced Research Projects Agency, Air Force Research Laboratory, Air Force Material Command, USAF under agreement number F30602-00-2-0546, by the Air Force Office of Scientific Research under grant number F49620-00-1-0072 and by the US Army War College.