Why AI Starves Without Human Creativity: Measuring and Renewing the Epistemic Commons

Abstract

Generative AI appears to the user as one intelligence, although its competence draws on traces left by populations. We call the machine Shannon’s demon. It can starve amid an abundance of text: when human effort declines and outputs of the same models fill the corpus, volume grows while variety and independence contract. One million observations whose errors correlate at 0.5 give the precision of two independent ones. Five users who ask the same model for the strongest objection each receive a Socratic conversation, and the population receives one new centre. Every answer carries two bills: the provider meters the computation; the effect on the knowledge that feeds future answers stays unpriced. The paper sets out what must be measured to say that the demon is starving, how a Socratic operator works for a population, and a deposit-funded pool with delayed rewards and plural validation that pays for the difference.