HHMI commits $500M to AI-guided biology, admits the tools aren't there yet

$500 million over 10 years. The release itself says it plainly: scientists do not yet know whether AI can deliver reliable mechanistic understanding of biology at this depth.

On June 15, Janelia Research Campus, part of Howard Hughes Medical Institute (HHMI, a private U.S. biomedical institute), announced two connected programs: building a mechanistic explanation of how a vertebrate brain generates behavior, with the transparent fish Danionella as the model organism, and AI-in-the-Loop, where AI helps plan and launch experiments as new data comes in. The campus was built for projects hard for a typical university lab to sustain over decades: new microscopes, sensors, neural maps, genetic tools, large open datasets.

Danionella. Slightly larger than a grain of rice, transparent in adulthood. In the standard model organism, zebrafish larvae, transparency exists only early in life, when behavior is still limited. Adult Danionella forages, learns, interacts with other fish, and engages in mating behavior, all while researchers can see most of its vertebrate brain. The goal: a mechanistic picture connecting specific cells and circuits to specific actions.

AI-in-the-Loop. In HHMI's version, AI reads incoming data, builds hypotheses, proposes the next experiment, and simulates options in advance. It also spots unexpected patterns and updates predictions after new measurements. In the mature version, part of the experimental cycle will run as a chain: the model proposes, the lab tests, the model revises the plan. The motive is practical: modern biology drowns in measurements (microscopy, neural activity, behavior, molecular markers, genetic lines), and manually working through thousands of condition combinations stalls research.

Janelia states the constraints openly. Danionella still needs genetic tools, imaging systems, behavioral assays, and models built; the first years will refine methods on drosophila and zebrafish larvae. The AI@HHMI budget is $500 million over 10 years. It includes protein design for fluorescent tags, cryo-electron tomography, metabolic sensors, RNA structures, robotic drosophila work, and zebrafish models. Janelia's new program looks like the place where those pieces come together into a single research process.

If AI-in-the-Loop actually shortens the hypothesis-to-experiment cycle from months to days, this kind of infrastructure spreads beyond neuroscience. Any complex biological problem requires the same loop: observe the system, choose an intervention, test the response, update the model, choose the next experiment. The institute writes it plainly: they do not yet know whether AI can deliver reliable mechanistic understanding at this depth. The central question is how quickly biology can turn measurements into working models.

Open the HHMI page on Eternal Search

Sources
t.me
t.me
hhmi.org
ai.hhmi.org