Stable Menus of Public Goods: AI-Enabled Progress

Problem
This work addresses the gap in understanding how AI can enhance economic research workflows, specifically in the context of public goods, as outlined in the EC 2025 paper “Stable Menus of Public Goods.” The authors investigate the effectiveness of various AI-driven methodologies, including the role of human intuition in prompts and the impact of multi-turn interactions. This research is presented as a preprint and has not undergone peer review.

Method
The study employs a series of experiments to evaluate three primary questions regarding AI workflows in economic contexts. The first question examines whether incorporating human intuition into prompts improves the performance of a large language model (LLM). The second question assesses the benefits of multi-turn interactions, where the model engages in iterative dialogue to refine outputs. The third question compares the LLM’s effectiveness against that of a first-year PhD student using an unpublished manuscript authored by the senior researchers. The LLM’s architecture specifics are not disclosed, but the experiments leverage a standard LLM setup for economic problem-solving tasks.

Results
The findings indicate that prompting with human intuition significantly enhances the LLM’s “taste,” leading to more relevant and contextually appropriate outputs. Multi-turn interactions were shown to be beneficial, particularly when the workflow encourages ambitious problem-solving steps. In the comparative analysis, the LLM was found to be slightly less effective than the first-year PhD student, suggesting that while LLMs can assist in generating ideas, they may not yet surpass human expertise in nuanced economic reasoning. Specific performance metrics are not provided, but the qualitative assessments suggest a notable difference in output quality.

Limitations
The authors acknowledge several limitations, including the lack of a comprehensive evaluation framework for quantifying the effectiveness of the LLM versus human performance. The reliance on a single unpublished manuscript for comparison may introduce bias, as it does not represent a broad spectrum of economic research capabilities. Additionally, the study does not explore the scalability of these workflows or their applicability across different economic contexts beyond public goods.

Why it matters
This research has significant implications for the integration of AI in economic research, particularly in enhancing the efficiency and creativity of problem-solving workflows. The insights regarding the importance of human intuition and multi-turn interactions can inform future AI applications in economics, potentially leading to more robust models and methodologies. As the field evolves, understanding the interplay between human and machine contributions will be crucial for advancing economic theory and practice, as highlighted in the ongoing discourse on AI’s role in research, as published in arXiv cs.AI.