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

Abstract:

Data aren’t just the fuel of artificial intelligence. Data granularity, frequency, and quality determine the feasibility and performance of AI algorithms. In the context of the generalized second-price auction used to sell internet search ads, we conduct simulated experiments with asymmetric bidders competing through Q-learning algorithms under different information structures on rival bids. We find that when less detailed information is available to train algorithms auctioneer revenues are substantially and persistently higher. This underscores the incentive for the digital platforms designing data-sharing policies to distort data flows to their advantage by strategically obfuscating data.