Large Language Models as Voting Mechanisms -- by Raša Karapandža, Yaw Nyarko

A study modeled large language models as voting rules where token probabilities reflect vote shares from training data. This structure produces Condorcet cycles and intransitive recommendations, observed in millions of pairwise stock comparisons on ChatGPT-4o. Correcting these circular preferences after generation is generally impossible.

Summary written by AI Business Lens with an AI model from the article at nber.org. It is not the article, and the publisher has not reviewed it. For publishers.

Read the original at nber.org   Generate teaching or study materials

Instructors get discussion guides, assignments, and mini-cases. Students and readers get a plain summary, points to raise, an exercise, and a self-check. All built from the full article. A free account saves stories and gives you three generations. Stories stay on the brief for 14 days. After that this page keeps the link to the original.

More in Economics

• Presentation: Building Reusable Evaluation Frameworks for Agentic AI ProductsInfoQ · October 5, 2026
• We’ve tracked $80M of spending on almost 170 AI-generated political ads this year – it’s weirdThe Conversation — AI · October 5, 2026
• Does the gender pay gap extend to AI? Research finds female AI agents 'paid' lessPhys.org — Technology · October 5, 2026
• The AI boom is making the world’s cheapest smartphones disappearRest of World · October 5, 2026
• Financing the AI Buildout -- by Stijn Van NieuwerburghNBER Working Papers · October 5, 2026