Simons Institute: Prediction, Incentives & Causal Intervention

This lecture introduces "triage scores" as a counterfactual risk assessment instrument, addressing how intervention affects outcomes in high-stakes decision-making beyond mere prediction.

This session presents a large-scale RCT on LLM-powered AI tutors, exploring personalized practice sequences using a POMDP framework for student mastery inference in Python certification.

This talk examines non-parametric causal inference in dynamic thresholding designs, addressing the limitations of naive regression-discontinuity analysis when temporal dynamics are present.

This lecture explores human-AI collaboration in decision-making, leveraging Aumann’s classical agreement theorem to understand how humans and AI models can achieve better outcomes together.

This presentation discusses the embedding of human values in clinical algorithms, examining how both traditional equations and large language models influence diagnostic reasoning and management.

This lecture investigates how predictive risk models in child welfare, when acting as policy, learn "risk" from administrative records shaped by bureaucracy, resources, and discretionary judgment.

This session proposes a revealed-preference approach to evaluate probabilistic decision-making in large language models, applying a decision-theoretic framework to high-stakes applications like medical triage.

This talk compares disparate impact and UDAAP legal frameworks for evaluating algorithmic discrimination, formalizing their application in a simulated lending setting to assess harms.

This lecture analyzes RCTs evaluating service interventions under capacity constraints, using queueing theory to model "operational dosage" and its impact on treatment effect estimation.

This talk revisits the debate on standardized tests in admissions by analyzing a dataset of master's program applications, finding GRE scores improve first-year grade predictions.

This lecture explores individual gaps in prediction generalization, focusing on challenges in applying predictive models universally across diverse populations or contexts.

This presentation examines how modeling decisions in statistical profiling impact fairness, analyzing regression and machine learning approaches using German labor market data.

This lecture introduces a dynamic model of firm-applicant interactions to identify mechanisms driving long-term statistical discrimination, highlighting the role of initial evaluation ability.

This session discusses decision-theoretic foundations for human-AI collaboration, presenting measurement frameworks rooted in statistical decision theory and information economics for team performance evaluation.

This lecture delves into the principles and methodologies of evaluation design, focusing on critical considerations for assessing interventions in social systems.

This talk presents an informational intervention designed and deployed in collaboration with NYC Public Schools, aiming to mitigate disparities in the complex high school match process.

This lecture examines historical experiments in technological federalism, tracing the automation of the US welfare state and its implications for social policy and administration.

This presentation discusses applying machine learning for predicting and prioritizing the allocation of scarce societal resources, drawing on theories of local justice and street-level bureaucracy.