Artificial Intelligence/Machine Learning Economics: Transparency, Competition, and Collusion

 Machine learning (ML) algorithms are being increasingly applied to decision-making processes with far-reaching impacts extending to employment, access to credit, and education. While ML algorithms have shown great predictive power in various business applications, there are rising questions about the economic and social consequences of algorithmic decision-making, and increasing calls for algorithmic transparency. In my dissertation, I study the impact of AI/ML on economic systems on several important aspects: transparency, competition, and collusion. 

Chapter 1, co-authored with Professor Param Vir Singh, Yan Huang, and Stefanus Jasin, examines the economic implications of algorithmic transparency in a hiring context. Specifically, it answers the following research question: Should firms that apply machine learning algorithms in their decision-making make their algorithms transparent to the users they affect? Despite the growing calls for algorithmic transparency, most firms have kept their algorithms opaque, citing potential gaming by users that may negatively affect the algorithm’s predictive power. We develop an analytical model to compare firm and user surplus with and without algorithmic transparency in the presence of strategic users. We identify a broad set of conditions under which making the algorithm transparent actually benefits the firm. By contrast, users may not always be better off under algorithmic transparency. These results hold even when the predictive power of the opaque algorithm comes largely from correlational features and the cost for users to improve them is minimal. These results suggest that firms should not always view manipulation by users as bad. Rather, they should use algorithmic transparency as a lever to motivate users to invest in more desirable features.

Chapter 2, co-authored with Professor Param Vir Singh and Yan Huang, studies strategic information revelation in the algorithmic lending space, and its impact on competition and welfare. Financial lenders’ opaque use of algorithms to screen unsecured credit applicants, coupled with high borrower uncertainty and search costs, can lead to sub-optimal credit decisions by many borrowers. Although some lenders facilitate informed decision-making for borrowers by providing personalized pre-approval probabilities, not all lenders do so.  In this study, we examine how competition among lenders influences their decision to disclose approval odds to borrowers via pre-approval tools. Our findings suggest that competitive pressures, particularly in cases where lenders’ algorithms are accurate, can undermine disclosure incentives. Lenders strategically employ asymmetric disclosure of pre-approval outcomes to reduce competition and differentiate their products. We demonstrate that borrower surplus is maximized when both lenders provide pre-approval tools and minimized when neither lender does so. However, mandating all lenders to provide personalized pre-approval outcomes may not necessarily enhance borrower surplus.

Chapter 3, co-authored with Professor Param Vir Singh, Yan Huang, and Kannan Srinivasan, studies algorithmic pricing and its impact on competition and collusion. The increasingly popular automated pricing strategies in e-commerce can be broadly categorized into two forms: simple rule-based algorithms, such as undercutting the lowest price, and more sophisticated artificial intelligence (AI) powered algorithms, like reinforcement learning (RL). RL algorithms are particularly appealing for pricing due to their ability to autonomously learn an optimal policy and adapt to changes in competitors’ strategies and market conditions. Despite the common belief that RL algorithms hold a significant advantageover rule-based strategies, our extensive experiments, conducted under both a canonical Logit demand environment and a more realistic non-sequential search structural demand model, demonstrate that when competing against RL pricing algorithms, simple rule-based algorithms can lead to higher prices and benefit all sellers, compared to scenarios where multiple RL algorithms compete against each other. Theoretical analysis in a simplified setting yields consistent results. Our research sheds new light on the effectiveness of automated pricing algorithms and their interactions in competitive markets, providing practical insights for retailers in selecting appropriate pricing strategies.