PhD Thesis Defense: Bruno Miranda Henrique

"Dynamic trust calibration"

Abstract

Trust calibration between humans and Artificial Intelligence (AI) is crucial for optimal decision-making in collaborative settings. Excessive trust can lead users to accept AI-generated outputs without question, overlooking critical flaws, while insufficient trust may result in disregarding valuable insights from AI systems, hindering performance. Despite its importance, there is currently no definitive and objective method for measuring trust calibration between humans and AI. Current approaches lack standardization and consistent metrics that can be broadly applied across various contexts, and they don't distinguish between the formation of opinions and subsequent human decisions. This thesis brings a novel and objective method for dynamic trust calibration, introducing a standardized trust calibration measure and an indicator. By utilizing Contextual Bandits — an adaptive algorithm that incorporates context into decision-making — the indicator dynamically assesses when to trust AI contributions based on learned contextual information. This indicator is evaluated across three diverse datasets, demonstrating that effective trust calibration results in significant improvements in decision-making performance, as evidenced by 10 to 38\% increase in reward metrics. Motivated by those results, human study experiments were conducted with a statistical sample of 500 participants, grouped into different treatments to evaluate human-only and human-AI performance, as well as the improvements made by making the indicator available in real-time decision-making. Another treatment also included a decision agent based on trust calibration. Experimental results show significant performance increase when the trust calibration indicator is available. Improved performance is also achieved with the decision agent availability. Results support the acceptance of external trust calibration tools by humans, highlighting risks and challenges when humans incorporate trust signals into decisions. These findings not only enhance theoretical understanding but also provide practical guidance for developing more trustworthy AI systems supporting decisions in critical domains, for example, disease diagnoses and criminal justice.

Thesis Committee

• Eugene Santos Jr. (chair)
• George Cybenko
• Wesley Marrero
• Hien Nguyen (University of Wisconsin - Whitewater)

Contact

For more information, contact Thayer Registrar at thayer.registrar@dartmouth.edu .