Advances in machine learning have powered AI agents to compete on the same level or even outperform humans on several challenging benchmarks. Such agents hold the promise to uplift our ways of living in various aspects. But to turn them into fruitful innovation, we need to be able to harness their capabilities and integrate them into our society without causing social turbulence. Achieving this goal requires building AI agents that can communicate naturally and effectively with humans. In this talk, I will present my two recent lines of work that propose new human-AI communication paradigms. The first line of work enables agents to learn from language descriptions of their activities (without any rewards and demonstrations of target behaviors). The second focuses on teaching agents to understand and convey their uncertainties and needs to leverage human assistance. I will also briefly present ideas for unifying interactive learning algorithms into a single human-centered, cognition-grounded framework.
Zoom: https://umd.zoom.us/j/98806584197?pwd=SXBWOHE1cU9adFFKUmN2UVlwUEJXdz09
(passcode if needed: clip)
Khanh Nguyen is a PhD candidate at the University of Maryland, College Park, where he receives guidance from Prof. Hal Daume III. His research is mainly on the topics of grounded interactive learning, with the goal of building AI agents that extend the capabilities of humans. He is currently focusing on teaching AI agents to communicate naturally and effectively with humans to better assist them. He obtained his bachelor's degree in Computer Science at the University of Massachusetts, Amherst.