LEMAS Seminar by Professor Alireza Fallah (Rice University) on Alignment from Response Times

Title: Alignment from Response Times: Learning Heterogeneous Preferences Beyond Binary Choices Abstract: Human feedback is often reduced to binary choices: one option is preferred to another, one answer is ranked above another, or one policy is selected over a competing alternative. This talk argues that such choice-only data can discard economically and statistically important information. Building on a general framework for estimating preferences from both choices and response times, I will show how response-time data can sharpen preference estimation, deliver fast convergence rates under drift-diffusion models, and improve the estimation of economically meaningful parameters in applications such as intertemporal choice. I will then connect this methodology to a central problem in AI alignment: learning a representative reward model from heterogeneous, often anonymous human labelers. In this setting, pooled binary feedback can make population-average preferences unidentifiable, leading to systematic distortions in the learned policy. I will show that response times provide a simple, low-cost additional signal that can restore identifiability and enable consistent estimation of average preferences, even when each labeler contributes only a single comparison. Bio: Alireza Fallah is an Assistant Professor in the Department of Computer Science at Rice University. In Spring 2026, he was an Apple Research Fellow at the Simons Institute for the Theory of Computing as part of the program on Federated and Collaborative Learning. He earned his Ph.D. in Electrical Engineering and Computer Science from MIT in summer 2023, working with Asu Ozdaglar and Daron Acemoglu. Prior to joining Rice, he was a postdoctoral researcher at UC Berkeley, hosted by Michael Jordan. He also spent Fall 2023 as the Gamelin Postdoctoral Fellow at the Simons Laufer Mathematical Sciences Institute, formerly MSRI, where he participated in the Mathematics and Computer Science of Market and Mechanism Design program. His research spans machine learning theory, market and mechanism design, game theory, optimization, and responsible AI, including privacy, fairness, and AI alignment. He has received several awards and fellowships, including an honorable mention for the ACM SIGecom Doctoral Dissertation Award, the Ernst A. Guillemin MIT M.Sc. Thesis Award, the Gamelin Postdoctoral Fellowship, the Apple Scholars in AI/ML Ph.D. Fellowship, the MathWorks Engineering Fellowship, and the Siebel Scholarship.