Ph.D. Candidate in Economics, George Mason University
Interdisciplinary Center for Economic Science
I am on the 2026–2027 job market.
I am a Ph.D. candidate in Economics at George Mason University and a graduate research assistant at the Interdisciplinary Center for Economic Science. I study human and AI decision-making using economic theory, behavioral experiments, and meta-analysis.
My job market paper asks whose welfare AI prioritizes when making economic decisions on people’s behalf, and shows that the answer depends on the decision perspective a task induces. My other research connects deliberate strategy search with learning from feedback to explain how people choose and reset strategies. I also develop reproducible methods for AI-assisted evidence synthesis, and I independently created and launched AI Behavioral Profile, a bilingual website comparing 10 AI models across 30 behavioral measures.
Fields Behavioral and Experimental Economics · AI Delegation and Decision-Making · LLM Behavior · Reproducible Evidence Synthesis
When people delegate economic decisions to AI, whose welfare does it prioritize? Using our reproducible AI-assisted evidence synthesis workflow (RAES), we synthesize 757 comparisons from 54 studies of economic games. AI shares and cooperates more than humans on average and responds more strongly to unfair treatment, but these differences vary widely across studies. We propose that this variation reflects the decision perspective induced by a task: acting from one participant's position (the self-interested perspective) or evaluating outcomes as a third party (the third-party perspective). Experiments with GPT-4o and GPT-5.5 support this explanation: changing whom AI is instructed to represent changes who benefits. To quantify the difference between perspectives, we estimate a welfare model using separate allocation and ultimatum tasks. Under the self-interested perspective, the weight on the participant AI represents is 1.00 for GPT-4o and 0.96 for GPT-5.5; under the third-party perspective, the weight on that same participant falls to 0.29 and 0.39, with the remainder on the other participant. With these estimates fixed, we identify the perspective from recovered study instructions and predict AI choices and their differences from human choices in the studies' one-shot, two-player games. Predictions using the identified perspective are more accurate than those using the opposite perspective: mean absolute prediction error for AI choices is 30 percent lower with the GPT-4o estimates and 60 percent lower with the GPT-5.5 estimates. How a task is described can therefore change whose interests a delegated AI serves.
How do people reset a strategy, and how does that choice shape what they do next? The framework connects deliberate strategy search at the start of a period with learning from feedback within it, tested in multi-period public goods experiments.
Develops a reusable workflow linking researcher-defined rules, AI execution, independent AI audits, deterministic computation, and offline reproduction.
Zhu, Q. (2021). Confirmation Bias and Gambler’s Fallacy Effect with Bayesian Method. In Proceedings of the 4th International Conference on Economic Management and Green Development (pp. 408–414). Springer. doi:10.1007/978-981-16-5359-9_47
A bilingual website comparing 10 AI models across 6 categories and 30 behavioral measures, with stability checks. Tasks adapt classic experiments from economics and established instruments from psychology; every measure documents its task design, scoring rule, and sources, so results can be traced and replicated. I independently developed the research framework, behavioral tasks, evaluation workflow, and website.
An open-source workflow for reproducible AI-assisted evidence synthesis. It connects researcher-defined rules, AI execution, independent AI audits, and deterministic computation, with templates and tools that make decisions traceable and results reproducible from saved records.
Student evaluations (Summer 2026): 4.7/5 for instructor preparation and course organization, and above the Economics department average on all 14 items (6 of 21 students responded).