EECS Colloquium: Explainable human-AI interaction for sequential decision support

About the event

Abstract:
For AI-powered systems to be truly successful in practical and everyday scenarios, it is not just enough for these systems to generate optimized decisions, but they need to be capable of working and collaborating with users from all walks of life. One major requirement for developing such systems is the need to imbue them with the ability to effectively model the expectations of their users, and to be capable of explaining their decisions and the rationale behind them in intuitive terms when such expectations cannot be met. In this talk, I will be describing some of the work  I have done in these directions, particularly to generate explanations in various sequential decision-making settings. I will discuss my solutions to some of the core challenges related to explanation generation, namely knowledge, inferential capability, and vocabulary asymmetry between the user and the decision-maker. I will also describe some of the use-case scenarios that have leveraged these methods and end the talk by discussing how we can extend these principles to enable people to provide advice and preferences to the AI system.

Bio:
Sarath Sreedharan is a Ph.D. student from Yochan Lab at Arizona State University. His core research interests lie in designing human-aware decision-making systems. His research has been published in various premier research conferences, including, AAAI, ICAPS, IJCAI, AAMAS, IROS, HRI, and ICRA, and in journals like AIJ, and has, to date, garnered over a thousand citations. He presented tutorials on his research at various forums and is the lead author of a Morgan Claypool monograph on explainable human-AI interaction, to be published this fall. A system built around applying the explanatory techniques he developed in the context of debugging goal-direct dialogue agents won the Best System’s Demo and Exhibit award at ICAPS-20. Before his Ph.D., he completed his MS (Computer Science) from ASU, where he was awarded the CSE Outstanding Masters Student award for the academic year 2016-2017. He has reviewed for and helped organize multiple premier AI conferences.  At  AAAI-20, he was recognized as an Outstanding Program Committee Member (one of 12 recognized).