The SCI Institute

Three SCI Faculty Among Recipients of RAI Seed Grants

The University of Utah One-U Responsible Artificial Intelligence Initiative (One-U RAI) will award its first round of seed grants, totaling $386,855, to nine research teams advancing AI tools and uncovering new insights on AI’s societal impacts. Projects will empower patients and their families, shed light on AI’s role in the classroom, and help people better understand environmental exposures that harm their health.

From a highly competitive pool of 40 proposals, One-U RAI committee members selected nine projects that reflect a balance of disciplines and initiative thematic areas—environment, health care and wellness, and teaching and learning. Awardees span 17 departments from across campus.

Bei Wang, Rob MacLeod, and Tolga Tasdizen were among the recipients.

Conversational AI Agents as Personalized Teaching Assistants

PI: Bei Wang Phillips, Computing
Co-PI: Rob MacLeod, Biomedical Engineering
Thematic Area: Teaching and Learning

Phillips and MacLeod will develop EduPilot, a conversational agentic AI teaching assistant, to integrate real-world computational workflows into biomedical engineering education. By translating natural language prompts into multi-step data analysis and visualization tasks, EduPilot lowers technical barriers while reinforcing conceptual understanding and scientific reasoning. Deployment in two courses will advance responsible AI integration in STEM education, strengthen workforce preparation, and promote reproducible computational practice.

Responsible AI for Rare Clival Tumors: Exact Patient-Level Unlearning Without Retraining

PI: Tolga Tasdizen, Electrical and Computer Engineering
Co-PI: Tyler Richards, Neuroradiology
Thematic Area: Health Care and Wellness

Tasdizen and Richards will develop an unlearning-capable MRI model to improve diagnosis of rare skull-base tumors while making patient consent withdrawal practical and auditable. By separating a shared backbone from removable patient-specific “plug-in” updates, the model can exclude withdrawn data without full retraining. The approach will be evaluated for tumor segmentation, classification accuracy, and measurable privacy protection, establishing a framework for adaptable and accountable medical AI.