AI is creating a 'contamination trap' where it looks like we're getting smarter, but the actual frontier of human knowledge is shrinking.
March 25, 2026
Original Paper
Innovation-Eroding Growth under AI Contamination and the Cost of Prevention
SSRN · 6454458
The Takeaway
The paper reveals that 'data rot' from AI makes standard economic indicators like GDP look healthy even as research quality collapses. Crucially, it finds that preventing this decline today is 50 times cheaper than trying to fix it in a decade.
From the abstract
AI-generated content is degrading the data commons on which frontier research depends. We develop a semi-endogenous growth model in which data quality evolves endogenously and show that contamination beyond a composition threshold traps innovation in self-reinforcing decline. Under laissez-faire, the economy enters a regime of innovation-eroding growth: aggregate output continues to rise-buoyed by derivative production-while the frontier knowledge base contracts, so that standard macroindicators