Title: The AI Precision Paradox: When Better Signals Make the Crowd Less Wise

Speaker: Guo Junjie, Associate Professor, School of Finance, Central University of Finance and Economics. His research covers monetary and fiscal policy and macro-finance. He has published more than 20 papers in top domestic and international economic journals and presided over General and Youth Projects of the National Natural Science Foundation. He works as an anonymous reviewer for many Chinese and English journals and NSFC projects, and won the second prize of the 18th Beijing Philosophy and Social Sciences Outstanding Achievement Award as a team member.

Abstract: AI is becoming a shared information–processing technology. This paper studies when a technology that improves each user's signal precision might nevertheless reduce the social value of information aggregation in some cases. Agents choose between costly independent research and a cheap AI–assignal that is individually precise but contains idiosyncratic noise and a common model error. AI adoption raises private signal precision, but also increases cross–agent error correlation, weakening the diversification logic behind the wisdom of crowds. Information diversity acts as a public–good–like input into aggregation: each atomistic agent treats it as given, but collectively agents determine whether errors diversify away. In the single–model benchmark, the key welfare object is the common–error share of AI noise. When this share is low, AI is socially beneficial; when moderate, AI improves welfare but is over–adopted; when high, individually superior AI signals become collectively welfare–reducing, generating a monoculture trap. The model also delivers an AI precision paradox: better AI can make the trap more likely by drawing more agents onto the same correlated signal. A dynamic extension shows that adoption can improve future AI quality through learning, but can also degrade the deployed information environment when AI–mediated content is recycled without sufficient filtering. Independent information capacity is a depreciating stock, so AI reliance can erode the outside option that would otherwise discipline adoption. Finally, with heterogeneous AI models, competition reduces concentration but need not restore informational diversity when models share data sources, or deployment–layer feedback. The paper provides a framework for evaluating when AI adoption enhances aggregation and when it undermines the independent information that makes aggregation valuable.

Date & time: 09 June 2026, 10:00 - 12:00

Venue: B321, Zhixin Building, Central Campus, SDU