Algorithms and User Behaviors

Abstract

This dissertation delves into the burgeoning and critical field of artificial intelligence (AI), focusing on the increasingly intricate interaction between sophisticated algorithms and complex user behaviors. In an era of digitalization and AI, understanding the interaction between algorithms, particularly those grounded in AI, and user behavior, has become a necessity rather than a luxury. The vast and diverse applications of AI algorithms have now permeated various aspects of human life, predicting and affecting behaviors in ways previously unforeseen. Yet, the extent to which AI algorithms predict, shape, and in turn are shaped by, user behaviors remains a vastly unexplored frontier. As such, this dissertation seeks to elucidate these interactions, bridging the gap in knowledge and contributing to the optimal design and effective use of AI algorithms in various industries. The first essay studies how emotion-aware algorithms can be developed to advance predictions of user behaviors. After developing an algorithm for multi-dimensional emotion detection in texts and conducting predictive analyses, the chapter presents a laboratory experiment study, underscoring the causal mechanisms through which the algorithm generates predictive power to user behaviors. The second essay combines video analytics algorithms and econometric models to understand user attention and clicking behaviors to video advertising. After empirically demonstrating that algorithms can be used to predict and understand user behaviors in various business contexts, the third essay focuses on the scenario where algorithms may shape user behaviors, particularly strategic behaviors. By formulating a game-theoretical model, it identifies the conditions conducive for adopting emotion-aware AI and determines the optimal allocation policies. The essay also delves into the welfare impacts of emotion AI on individuals, organizations, and society at large. The results suggest that a stronger AI is not always socially desirable and necessitates regulation on data-driven allocation. Moreover, it outlines scenarios in which AI adoption can be more profitable than employing human labor for emotion recognition and resource allocation.