Talk title: Human-Centered AI: From Clinical Education Support to Enabling End-user Inspection of AI
Modern generative AI offers significant capabilities, yet integrating these technologies with the practical needs of domain experts and end-users is challenging. Additionally, end-use issues like social bias, stereotyping, and limited explainability can undermine trust, safety, and AI adoption, highlighting critical hurdles in AI deployment in real-world settings.
In this talk, I will present my work in Human-Centered AI, focusing on two core projects. First, I will explore the use of unsupervised, explainable Human-AI collaboration techniques. These techniques are applied to discover clinically relevant categorizations of unstructured verbal feedback during live surgery. This work represents a key effort into applying AI in clinical education developed alongside clinicians. Second, I will showcase efforts to enable end-users to examine social biases in large language models (LLMs) through user-friendly interfaces, an initiative crucial for enhancing transparency and trust in AI. I will also highlight my mentorship of undergraduate students who have significantly contributed to these projects. Together, these efforts demonstrate a blend of technical innovation and user-centric design in my AI research approach.