Seminario Auditing AI Algorithms: Evidence from Field Experiments

The seminar is reserved for the Department of Management’s community. Other interested colleagues can contact Eleonora Monaco (e.monaco@unibo.it).

Platforms increasingly deploy AI-driven estimation tools to inform users, yet no gatekeeper ensures this AI-powered information is not manipulated to attract users. We propose a methodology to audit intentional algorithmic bias and implement it through a field experiment on Uber in Hong Kong. Analyzing over 500 rides, we randomize ride contexts to vary the incentives for information manipulation. By comparing pick-up time estimates with travel-time estimates, both subject to similar technical forecasting challenges such as traffic conditions, weather, and driver variability, we isolate strategic distortion from genuine forecasting error. We document a systematic 43% underestimation of pick-up times, with no comparable bias in travel-time estimates. The bias intensifies when riders face fewer alternatives and during peak demand periods, yet attenuates when riders appear time-attentive and thus more likely to detect manipulation. Our results contribute to the growing debate on AI regulation and underscore the urgent need for auditing AI algorithms across any platform that deploys AI-driven estimation, including ride-hailing, food delivery, e-commerce, and other digital platforms, to attract and retain users.

Dr. Cao is the Director and Founding faculty of the AI Initiative for Capital Market Research and holds the position of associate professor (with tenure) at the Robert H. Smith School of Business, University of Maryland. Additionally, he is an affiliated professor at Harvard Business School (D^3 Institute). Dr. Cao's research work has gained prominence in respected media outlets such as the Financial Times, CNBC, Bloomberg, The Guardian, Quartz, and IR Magazine. His research papers, some co-authored with PhD students, have received several best paper awards, including the Fama-DFA Prize from Journal of Financial Economics for From Man vs. Machine to Man + Machine: The Art and AI of Stock Analyses and the Michael J. Brennan Award from Review of Financial Studies for How to Talk When Machines are Listening: Corporate Disclosure in the Age of AI. He has published in leading journals across finance, accounting, and computer science, including Journal of Accounting Research, Journal of Financial Economics, Journal of Financial and Quantitative Analysis, Review of Financial Studies, The Accounting Review, Contemporary Accounting Research, Management Science, and IEEE Computer. Dr. Cao serves as an Ad Hoc Editor at Contemporary Accounting Research, a Guest Associate Editor at Management Science, and a co-chair of FinTech and Machine Learning Conferences with the Review of Financial Studies, receiving 450 dual-option submissions.
Dr. Cao is deeply committed to helping business communities through his research. In addition to his role in founding and leading the AI initiative at the University of Maryland, he champions AI's impact in finance and accounting by delivering over 200 invited research talks at universities and national regulatory and policy-advisory agencies, including the Central Bank of Japan, the Central Bank of Thailand, and the U.S. Securities and Exchange Commission.