Abstract

Artificial intelligence (AI) is now an integral part of everyday decision making. These systems not only help us make choices but also learn from our behavior. Therefore, it is important to understand how our interactions with AI influence these systems. Current approaches often assume that the behavior used to train AI accurately reflects how people would naturally behave. However, people are often informed that their choices will be used for AI training, and as a result, may change how they would normally act. In this dissertation, we combine psychological and computer science perspectives to investigate the consequences of this awareness. We show that people alter their behavior when they know it will be used to train AI. These changes are shaped by contextual factors, including the extent to which people deliberate, how they engage with disclosure information, and their attitudes toward the individuals who will interact with the trained system. Consequently, people can embed their biases into AI, potentially influencing the systems' downstream decisions. These modifications can also affect both the individuals who provide training data and those who later interact with the deployed system. Behavioral patterns adopted during AI training persist after training has ended, suggesting that training-related behavior can become habitual. Additionally, when people learn how an AI was trained, they adjust their behavior based on their attitudes toward the group that trained it rather than the AI's behavior. This dissertation provides insights for both AI development and our broader understanding of human behavior. People rely on the same cognitive processes and social motivations when providing training data as they do in other decision-making contexts. As a result, the biases that shape human decisions can also influence the data used to train AI. Understanding how these mechanisms operate is essential for predicting when training data will diverge from people's baseline behavior and when biases are likely to become embedded in AI. These insights demonstrate how psychological theory can inform the design, evaluation, and deployment of AI systems, and highlight the importance of documenting all stages of the training and deployment pipeline to improve human-AI interactions.

Committee Chair

Wouter Kool

Committee Members

Alexandra Decker; Alvitta Ottley; Chien-Ju Ho; Jessie Sun

Degree

Doctor of Philosophy (PhD)

Author's Department

Interdisciplinary Programs

Author's School

McKelvey School of Engineering

Document Type

Dissertation

Date of Award

8-6-2026

Language

English (en)

Download

Share

COinS