Key Takeaways
- AURORA is an agentic AI system from Marinka Zitnik's Harvard Medical School lab that suggests hypotheses, designs experiments and analyzes results
- Three Pioneer Teams — NexTGen, PROMINENT and MATCHMAKERS — will each add the co-scientist with about $1.3 million in support
- Harvard stresses that humans remain responsible for all scientific judgment and decision-making
Harvard Is Giving Cancer Researchers an AI Lab Partner
Harvard Medical School has launched a test of a new kind of research partner: an AI system designed not just to answer questions, but to do the creative work of science alongside humans.
The system, called AURORA, was developed in the lab of Marinka Zitnik, an associate professor of biomedical informatics at Harvard Medical School. Unlike a chatbot, AURORA can pull together scientific literature, databases and other biological AI tools — including protein-folding and tumour models — along with real-time experimental results. From that, it suggests new hypotheses, designs experiments to test them (prioritizing time, budget or other factors the investigators set), and analyzes the results to propose what to try next. It also shows "reasoning traces," so scientists can see the rationale behind every output.
AURORA was launched November 5 with three Pioneer Teams, each receiving approximately $1.3 million to add the co-scientist to their work. NexTGen is adapting CAR T-cell therapy — proven in adult blood cancers — for pediatric solid tumours. PROMINENT is working to distinguish genetic mutations that merely accumulate in cells from the ones that actually trigger malignancy, which could one day allow intervention before cancer develops. MATCHMAKERS is tackling how T cells recognize cancer cells, toward more personalized immunotherapies.
"A key challenge for AI in science is understanding how it can best support researchers working on real-world problems," Zitnik said. Harvard is emphatic on one point: in the AURORA project, humans remain responsible for all scientific judgment and decision-making.
The Twikup Take
Most AI-in-science headlines are about a model beating a benchmark. This one is different: a co-scientist embedded in three real cancer labs, judged by whether experiments actually move faster and dead ends get cut sooner. If AURORA works under those conditions, it changes lab economics. If it doesn't, we'll learn that too — which is rather the point of the experiment.
Image: AI-generated illustration.