Scientific intelligence / MAGICS Lab
NORTHWESTERN UNIVERSITYWe develop AI that reasons about the world, designs interventions, and learns from experiments.
Our ambition: make scientific experience improve both what AI understands and what it can help us build.
The research agenda
Agents choose the next step. Models anticipate its consequences. Experiments test it against the world.
How can AI become better at research through experience? We study memory, reasoning, adaptation, and tool use to help agents improve their hypotheses, plans, and decisions from evaluated outcomes.
Explore the foundations →From genomes and cells to the changing universe. We develop representations of scientific objects and dynamics for prediction, uncertainty, and the design of interventions.
Explore the models →We develop systems that connect scientific agents with simulation, robotics, and automated experiments. The goal: a Design–Build–Test–Learn cycle that makes each measured outcome useful for the next decision.
Explore the systems →The compounding opportunity
Discovery creates knowledge. Invention puts it to the test. We study how AI can connect both, turning measured outcomes into better questions, designs, and experiments.
From discovery to invention →Shared intelligence. Verifiable experience.
Better evidence for the next decision.
From ideas to scientific infrastructure
Directed by Han Liu, Northwestern’s CFMG connects foundational advances in machine intelligence with science and engineering, bringing researchers together around models, reasoning, and discovery.
Explore the center ↗An NSF-supported cloud laboratory in development, connecting protein design, AI, robotics, and automated experimentation. Han Liu serves as Co-PI and AI lead.
Inside the initiative ↗Han Liu co-leads a scientific foundation-model project within the NSF–Simons SkAI Institute, learning from astronomical time-series data.
Explore the project ↗We welcome students, scientific collaborators, industry partners, and investors who share an interest in reasoning agents, scientific models, and experimental systems.
GitHub
Google Scholar
LinkedIn
CFMG