HCC: RI: Medium: Improving Human-AI Decision-Making Partnerships Through Shared Understanding

Artificial intelligence (AI) is becoming a key partner in critical human decision making — from diagnosing diseases to everyday driving. To ensure these partnerships work effectively, this project develops new ways to accurately measure and track how well both people and AI perform tasks over time. These assessments will help determine in what situations it is better for a human to take the lead in decision-making and in what situations the AI can be trusted. Equally important is creating AI systems that respect and reinforce human goals, like fairness, teamwork, and preserving a sense of personal control. By combining techniques from statistical modeling and cognitive science, this project will improve how people and AI collaborate, making these interactions not just effective but also aligned with human values and expectations. The research project is interdisciplinary in nature, building on Bayesian learning and cognitive modeling to systematically study and optimize human-AI interaction. It focuses on three core research activities: (1) developing Bayesian inference frameworks to dynamically evaluate the evolving abilities of both humans and AI agents; (2) creating adaptive optimization algorithms to manage decision policies under uncertainty, incorporating costs and constraints inherent in real-world human-AI collaboration; and (3) exploring human-centered aspects, integrating subjective metrics such as perceived fairness, teamwork, and agency into multi-objective optimization frameworks. Behavioral studies across various tasks—including image classification, natural language question answering, visual target tracking, and simulated navigation—will validate theoretical models and algorithms. Additionally, the project will create openly accessible datasets to support reproducibility and facilitate further research on collaborative human-AI systems. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.