Integrating Plan Generation, Execution, and Monitoring for Explainable, Normative, and Justified Agency

Abstract

Like humans, autonomous agents must operate in physical settings that involve changing situations and competing objectives. In this talk I describe PUG, an agent architecture that combines symbolic and numeric descriptions of states, associates utilities with symbolic goals, and uses mental simulation to evaluate plans. PUG interleaves plan generation, execution, and monitoring, with the latter detecting anomalies that lead to replanning. I demonstrate the architecture's behavior on scenarios from a continuous physical domain that involve tradeoffs among objectives and unexpected events.

In addition, I pose a new challenge for AI researchers -- developing intelligent agents that explain the reasons behind their activities and that attempt to follow social norms. In both cases, I describe the target abilities, examine some design alternatives, and outline plans for extending PUG to support them. Next I introduce the notion of justified agency -- explaining actions in terms of such norms. I conclude by proposing testbeds to evaluate these abilities and encouraging other researchers to tackle this challenge.

This talk describes ONR-funded work done jointly with Dongkyu Choi, Mike Barley, Ben Meadows, and Edward Katz. For details, see:

Langley, P., Choi, D., Barley, M., Meadows, B., & Katz, E. P. (2017). Generating, executing, and monitoring plans with goal-based utilities in continuous domains. Proceedings of the Fifth Annual Conference on Cognitive Systems. Troy, NY. http://www.cogsys.org/papers/ACS2017/ACS_2017_paper_17_LangleyEtAl.pdf

Langley, P. (2019). Explainable, normative, and justified agency. Proceedings of the Thirty-Third AAAI Conference on Artificial Intelligence. Honolulu, HI: AAAI Press. http://www.isle.org/~langley/papers/explain.aaai19.pdf

Bio

Dr. Pat Langley serves as Director of the Institute for the Study of Learning and Expertise and as Honorary Professor of Computer Science at the University of Auckland. He has contributed to artificial intelligence and cognitive science for more than 40 years, having published over 300 papers and five books on these topics. Dr. Langley developed some of the first computational approaches to scientific knowledge discovery, and he was an early champion of both experimental studies of machine learning and its application to real-world problems. He is the founding editor of two journals, Machine Learning in 1986 and Advances in Cognitive Systems in 2012, and he is a Fellow of both AAAI and the Cognitive Science Society. Dr. Langley's current research focuses on architectures for intelligent agents, abductive methods for plan understanding, and induction of explanatory scientific models.