Sold to the Highest Bidder: Can We Rethink Fairness in Targeted Advertising, or Are We Placing a Losing Bet?

Abstract

The pursuit of fairness in algorithms is often framed as a solvable mathematical problem, yet the concept of fairness itself has long been debated by philosophers, historians, and economists without resolution. If defining fairness were merely a statistical challenge, then computer scientists might have succeeded where others have failed. However, the real difficulty lies not in measuring fairness but in determining whose perspective defines it. Fairness is not an inherent property of an algorithm; rather it is a function of the relationship between the algorithm and the stakeholders involved. This thesis examines fairness in online advertising auctions, where algorithmic decision-making determines which users see which ads. Instead of treating fairness as a fixed standard, this work models different fairness definitions and evaluates their impact on ad distribution, efficiency, and stakeholder incentives. The modeling framework presented here does not seek to prescribe an "optimal" fairness solution but instead reveals how different fairness constraints shape online advertising outcomes, exposing trade-offs between fairness, efficiency, and revenue. By simulating ad auction environments under competing fairness criteria—such as statistical parity, equalized odds, and individual fairness—this thesis demonstrates how small changes in fairness definitions can yield vastly different systemic effects. Rather than positioning fairness as a purely computational objective, this work argues that fairness in ad auctions is fundamentally a policy decision, not a technical optimization problem. Through this modeling, it becomes evident that decision-makers—whether policymakers, companies, or regulators—hold the power to define and enforce fairness, shaping market outcomes through their chosen definitions. This thesis challenges the assumption that fairness can be universally quantified, instead advocating for transparency in the trade-offs that fairness interventions impose on different stakeholders. Algorithmic fairness is not about finding a single solution, it is about making explicit the hidden values embedded in digital decision-making and understanding that any notion of algorithmic fairness is ultimately a reflection of the human mind behind the machine: defining, building, and ultimately, making a decision.