Our weekly SRI Seminar Series welcomes Rediet Abebe, a Harvard junior fellow and Andrew Carnegie fellow whose research examines the interaction of algorithms and inequality. In this talk, Abebe will explore how algorithmic prediction techniques are emerging as a promising tool for efficiently allocating societal resources, and how new mathematical frameworks can better assess what types of predictions are necessary for efficient resource allocation in the context of inequality. This session will be moderated by Nisarg Shah.
Talk title:
“When does resource allocation require prediction?”
Abstract:
Algorithmic predictions are emerging as a promising tool for efficiently allocating societal resources. Fueling their use is the prevailing belief that accurately identifying individuals at the highest risk of adverse outcomes—such as loan defaults, poor health, or school dropouts—is a key bottleneck.
We challenge this assumption in this talk: We present findings from a multi-year study of Wisconsin's Early Warning Systems (DEWS). While DEWS effectively ranked students by dropout risk and may have modestly increased graduation rates, we demonstrate that these individual risk scores were not essential to achieving this outcome. A simple allocation mechanism using only aggregate school-level data would have sufficed. This finding is grounded in two pervasive truths about U.S. public schools: structural factors drive student dropout rates, and schools are highly segregated.
Generalizing these insights, we introduce mathematical frameworks to assess when accurate individual predictions are truly necessary for efficient resource allocation. In contexts where individuals are part of larger, structurally significant units like hospitals, neighborhoods, or schools, prediction-based allocations outperform simpler methods when between-unit inequality is low. We also explore the timing of interventions, weighing the trade-off between acting early on less precise predictions versus waiting for more accurate data. Here, too, inequality plays a critical role. In settings with high levels of inequality, early action based on noisy predictions is often more effective than waiting for better data.
These studies highlight the potential limits of improving allocation via predictive systems when inequality is high.
Based on joint work with: Tolani Britton, Moritz Hardt, Juan C. Perdomo, Ariel Procaccia, and Ali Shirali. This work relies on data and models provided by the Wisconsin Department of Public Instruction, and has been informed by discussions with Erin Fath, Carl Frederick, and Justin Meyer.
About Rediet Abebe
Rediet Abebe is a junior fellow at the Harvard Society of Fellows and an Andrew Carnegie Fellow. Her research examines the interaction of algorithms and inequality, with a focus on contributing to the scientific foundations of this area. Abebe has also co-founded numerous organizations, including the ACM Conference on Equity and Access in Algorithms, Mechanisms, and Optimization (EAAMO), the associated international research initiative, and Black in AI. Abebe is the recipient of numerous awards and honours, including the Hector Endowed Fellowship by the European Laboratory for Learning and Intelligent Systems (ELLIS), MIT Technology Fellows 35 Innovators under 35, the ACM SIGKDD Dissertation Award, and an honorable mention for the ACM SIGecom Dissertation Award. Abebe currently leads several large-scale evaluations of ML systems used in commercial, legal, and policy contexts.
About the SRI Seminar Series
The SRI Seminar Series brings together the Schwartz Reisman community and beyond for a robust exchange of ideas that advance scholarship at the intersection of technology and society. Seminars are led by a leading or emerging scholar and feature extensive discussion.
Each week, a featured speaker will present for 45 minutes, followed by an open discussion. Registered attendees will be emailed a Zoom link before the event begins. The event will be recorded and posted online.
Rediet Abebe