#SciFM24 at the University of Michigan
Program
Day 2
7:30 AM
Registration Desk Opens
Breakfast
8:00 - 9:00 AM
Opening Keynotes
9:00 - 10:30 AM
Keynote 1
From Digital to Physical: The AI Transformation
Chair: Ian Foster (Argonne National Lab / The University of Chicago)
Two opening keynotes framing the grand challenge: what does it mean for AI to move from pattern recognition on static data to active, closed-loop engagement with the physical world?
Keynote
Speaker: Hiroaki Kitano (Sony)
Keynote
Speaker: Michelle Li (Harvard University)
Coffee Break
10:30 - 11:00 AM
Industry Challenge Panels · Panels 1, 2, 3
Each panel pairs industry leaders with academic researchers to articulate specific problems that current AI approaches are not solving.
11:00 - 12:30 PM
Panel 1
Generative AI & the Industrial Transformation
Generative AI is beginning to reshape industry, yet many of the most consequential engineering challenges remain beyond the reach of today’s foundation models. Industrial systems operate under hard physical constraints, uncertainty, safety requirements, and complex multi-scale dynamics that current AI systems handle poorly. While modern AI excels at language and perception, it still struggles with physics-grounded reasoning, uncertainty quantification, causal modeling, and reliable decision-making in complex real-world environments.
This panel brings together leaders from industry and academia to examine what current AI gets wrong about industrial transformation — and what future Scientific Foundation Models (SciFMs) must do differently. The discussion will focus on the technical gaps preventing reliable deployment of AI in large-scale industrial settings, and on the new architectures, methods, and infrastructure required to enable trustworthy AI for engineering, science, and operations.
PANELISTS
Lunch
12:30 - 1:30 PM
1:30 - 3:00 PM
Panel 2
Biotech & Life Sciences: What AI Cannot Yet Do for Drug Discovery, Protein Engineering & Autonomous Biology
Biology is perhaps the domain where scientific foundation models have the highest stakes and the deepest unsolved problems. Despite remarkable progress in protein structure prediction, the landscape of drug discovery, enzyme engineering, and autonomous biological experimentation is littered with AI approaches that excel on benchmarks but fail in the lab. This panel directly confronts the gap between computational promise and experimental reality — and asks what a genuinely useful biological foundation model would need to do that current systems cannot.
Coffee Break
3:00 - 3:30 PM
3:30 - 5:00 PM
Panel 3
Embodied Intelligence: Closing the Perception–Action Loop in Science
From autonomous microscopes and robotic synthesis platforms to field-deployed environmental sensors and humanoid lab assistants. The challenge is not just locomotion — it is building AI systems that understand the scientific context of their actions and can adapt to the long tail of experimental variability.