Individual use of generative AI is creating a social trap that will eventually destroy the quality of the AI itself.
Model collapse occurs when AI systems are trained on data produced by other AIs, leading to a rapid decline in information diversity. Each user acting rationally to save time by using AI contributes to a polluted digital commons of synthetic data. This cycle functions like the Tragedy of the Commons, where individual benefits lead to a collective disaster for the entire ecosystem. As the pool of human-generated data shrinks relative to AI output, the models lose their ability to represent reality accurately. Maintaining high-quality AI requires a steady stream of original human thought that our current usage patterns are actively undermining.
Generative artificial intelligence reduces social welfare through model collapse
arXiv · 2604.21853
Generative artificial intelligence (genAI) is rapidly reshaping how knowledge and culture are produced and consumed. Yet generative models are vulnerable to model collapse: when trained on data generated by earlier versions of themselves, their outputs can lose diversity and accuracy. This creates a social dilemma, because delegating tasks to genAI can be individually beneficial in the short term even as widespread adoption degrades future model performance. Here we develop a parsimonious model