Artificial Intelligence & Data Obfuscation: Algorithmic Competition in Digital Ad Auctions

Abstract

Data are the key fuel of artificial intelligence and any change to the type and quality of available data has an impact on the type and performance of the feasible algorithms. We analyze the incentives that large digital platforms have to alter data flows to their advantage by strategically obfuscating data. We quantify this phenomenon in the context of digital advertising auctions through a series of simulated experiments where asymmetric bidders employ artificial intelligence algorithms to compete in Generalized second-price auctions. We find that when less detailed information is available to train algorithms, auctioneer revenues are substantially and persistently higher.

Keywords: Auctions, Procurement, Collusion, Data, Privacy

JEL Classification: C73, D82, D83, D18