Research & Writing

Working Papers

[JMP] Calibrated Coarsening: Designing Information for AI-Assisted Decisions, with Bnaya Dreyfuss [Latest version]
_{Abstract: Artificial intelligence (AI) is increasingly used to aid human decision-making across critical applications, but errors in human probabilistic reasoning (e.g., cognitive biases) often undermine its effectiveness. This raises the central design question of how to provide AI input to humans in a way that improves decision-making outcomes. We propose calibrated coarsening—partitioning the signal space into fewer cells at chosen thresholds—as a way to do so that (i) ensures humans retain final decision authority, (ii) modifies signals without deception, and (iii) adapts flexibly to diverse biases and contexts. Within an information disclosure framework, we derive an approximately optimal universal coarsened policy when the designer does not observe the decision-maker’s information. We then empirically demonstrate in a randomised experiment with professional loan specialists that coarsening AI signals at the theory-derived threshold significantly improves decision-making outcomes, over both the human-only (based solely on the loan application) and uncoarsened AI (assisted with continuous AI risk-score) benchmarks. We uncover substantial decision heterogeneity amongst loan officers and estimate a Bayesian hierarchical model to personalise coarsening policies, which we then test in a two-stage experiment.}

Encouraging Digital Wellness at Scale: Experimental Evidence from 13 Million Social Media Users, with Peter Hickman and Yulu Tang [Latest Version]
_{Abstract: We document the demand for, and effectiveness of, tools aimed at supporting user’s digital wellness through a large-scale natural field experiment orders of magnitude larger than any previous study. We partner with a leading social media platform to prompt 13 million active social media users to set Digital Wellness alarms—soft commitment devices in the literature and estimate a 4.8% take-up rate among this broad population. We consider and present evidence for why take-up may differ in this study from earlier small-scale studies in the literature: the prevalence of perceived self-control problems, the effectiveness of the alarm, and sensitivity to the timing and nature of the invitation to take up commitment. Moreover, we find evidence that higher demand in earlier studies is a consequence of lab experiments selecting participants with higher-than-average demand for commitment.}

Works-in Progress

Uneven Organizational Gains: GenAI’s Impact on Worker Specialization and Task Complexity

AI and Wellbeing, with Emanuel Schertz

Screen Time Limits and Youth Mental Health: Evidence from a Randomized Experiment, with Luca Braghieri, Sarah Eichmeyer, Matthew Gentzkow and Angela Yuson Lee

Understanding Privacy Preferences and Behaviour, with Andrew Kao and David Yang

Unreliable Electricity, with David Lagakos and Francesco Nuzzi

Case Studies

Riiid: Scaling AI Educational Services Globally, Harvard Business School Case 324-030, with John Jong-Hyun Kim and Nancy Dai [Published: Nov 2023]