Smitha Milli

University of California, Berkeley, Berkeley College, University of California System

Papers

Total Citations

H-Index

About

Smitha Milli is a leading researcher at the intersection of artificial intelligence, human-computer interaction, and value alignment. Her work focuses on the critical challenge of ensuring that AI systems learn and act in accordance with genuine human preferences, even when those preferences are complex or imperfectly expressed. Milli's most impactful contribution is the "reward-rational (implicit) choice" formalism (56 citations), a unifying framework for learning reward functions from diverse forms of human behavior, moving beyond simple feedback to interpret implicit choices. She has also made pioneering contributions to the "obedience problem" in robotics (19 citations), demonstrating that a robot blindly following orders can be harmful when humans are irrational, and instead arguing for robots that infer deeper human preferences. Her work on "pedagogic humans" (3 citations) further reveals how system designers often misspecify objectives, showing that people may actively teach robots, a nuance that must be accounted for in learning algorithms. Through this research, Milli is shaping a future where AI is not just capable, but also wise and aligned with our true, nuanced intentions.