Who Gets Paid When An AI Answers Your Question? | Parag Agrawal, Parallel Web Systems
Shapley values could fairly compensate web publishers when AI uses their content
“In web search, we can run simulations of if we did not have access to this URL or this domain or this collection of them, how would the agent perform?”
Parag Agrawal (ex-Twitter CEO, now Parallel Web Systems) proposes applying Shapley values to measure each web source's contribution to an AI answer, enabling fair payment to content creators. The approach uses eval-based simulation — withholding specific URLs or domains — to estimate counterfactual agent performance. This directly addresses the unsolved economic problem of how AI companies should compensate publishers whose content trains or grounds their models.