Building a Semi-Automated AI Pipeline for Applied Economics Research

Speaker: A/Prof David Smerdon

Affiliation: The University of Queensland

Location: Level 6 Boardroom (629), Colin Clark Building (#39), St Lucia Campus

Zoom: https://uqz.zoom.us/j/82603079317

Abstract: Large language models are already changing how economists search, code, write, and revise. But can they do more than assist with isolated tasks? This talk presents early work on building a semi-automated AI pipeline for applied economics research. I will begin with a brief overview of the rapidly moving frontier in automated AI research, including fully automated paper-generation systems and recent benchmarking efforts. I will then describe our current UQ project: a human-in-the-loop pipeline in which several frontier models independently propose, critique, and revise research designs across stages including literature review, data assessment, identification strategy, pre-analysis planning, estimation, interpretation, drafting, and simulated peer review. The talk will be deliberately candid: the system is deliberately not (yet) fully automated, and the most interesting lessons so far come from where it breaks. I will show results from pilot runs and benchmark-paper comparisons, discuss open problems around data access, econometric code generation, disagreement between models, and reproducibility, and close with broader predictions about how AI-assisted research will reshape both economics research and our profession.