Abstract

Abstract: I’ll discuss three projects related to understanding how people and AI-infused systems can and should interact. In the first, I’ll discuss AI communicating to people, in a shared environment, and how we can use highlighting and possible alternatives as a way to combat confabulations (aka hallucinations). In the second, I’ll discuss people communicating to AI systems, and how we can leverage language’s capability to describe the same behavior at multiple levels of abstraction. Finally, I’ll discuss people and AI interacting at the low level of predictive text systems, and how subtle differences in the behavior of the AI system can – or can not – change people’s behavior.

Bio:

Hal Daumé III is the Director of AIM , the AI Interdisciplinary Institute at Maryland. He is a Volpi-Cupal endowed Professor of Computer Science and Language Science at the University of Maryland, where he also leads TRAILS , an NSF & NIST -funded institute on Trustworthy AI. His research focus is on developing natural language processing systems that interact naturally with people, promote their self-efficacy, while mitigating societal harms. Together with his students and colleagues, he has received five best paper awards, a best demo award, and a test of time award. He has been program chair for the International Conference on Machine Learning in 2020 (together with Aarti Singh) and for the North American Association for Computational Linguistics in 2013 (together with Katrin Kirchhoff), and he was an inaugural diversity and inclusion co-chair at the Neural Information Processing Systems Conference in 2018 (with Katherine Heller).