Séminaire interne Nicolas Pasquier, Chargé de recherche INRAE

Mardi 10 novembre 2026 

Historical data as a barrier to entry in AI markets: The role of human 

Nicolas Pasquier

Résumé : The first AI tools to compete with human experts in giving advice were trained on the history of cases these experts had accumulated. Access to this history is now restricted, so new AI tools cannot use it. I ask whether such history is a barrier to entry. I build a Hotelling model in which two AI tools compete with human experts. The incumbent tool and the experts hold the history; the entrant does not. Both the tools and the experts can learn from the cases that users bring. When experts do not learn and are active, the history is not a barrier to entry. When experts learn and are active, it becomes one. The barrier, however, comes from the experts, not from the incumbent, which earns more when the entrant is active. Since users who switch to AI tools stop feeding the experts, I find that AI tools can also lower consumer surplus. Finally, I let the experts license their data to one firm. They choose the entrant, which removes the barrier, unless they earn enough from their users. In that case, they choose the incumbent or no firm. This rationalizes the contrasting strategies observed in practice: online forums licensed their data to incumbents, while agricultural cooperatives turned to new AI tools.

 
Le séminaire a lieu à 11h30, au GAEL