Errors, Efforts, and Earnings: AI-Driven Field Experiment in Suboptimal Labor Supply Decisions

This project evaluates whether AI-based decision-support tools can improve labor supply decisions and earnings among taxi drivers, a vulnerable and underserved segment of the urban gig economy. The researchers observe that in many cities, taxi drivers systematically over-queue at transportation hubs such as airports, despite earning significantly less per hour than city cruising. The researchers ask whether this behavior reflects rational preferences over effort and risk or distorted beliefs about earnings, and whether AI can alleviate these frictions.

Partnering with a leading AI mobility platform, the researchers implement a randomized controlled trial with 9,000 drivers in a 2×2 factorial design that independently varies access to personalized “Belief Calibration” feedback and real-time “Ability Improvement” tools (demand forecasts and routing). A staggered rollout allows the researchers to identify partial equilibrium effects while measuring general equilibrium spillovers.

Beyond its direct goal of raising earnings and reducing inefficient time allocation among taxi drivers, this project advances research on behavioral biases, technology adoption, and human-AI complementarity by testing whether “de-biasing” is a prerequisite for effective AI use. Positive results would offer a scalable template for governments and platforms seeking to use AI and administrative data to improve outcomes for gig workers in low- and middle-income countries.