Stanford launches five-year project to build AI models of the human brain

Stanford University is starting a five-year project to build AI models that can predict how the human brain will respond to different situations, its Wu Tsai Neurosciences Institute announced Monday.  The school did not say how much the project costs.

The Stanford Digital Brain Project will train what researchers call "large brain models."  They work like the foundation models behind chatbots such as ChatGPT or Claude, which learn broad patterns from huge amounts of text.  Here, the training data will be recordings of brain activity.  Stanford compares the goal to a weather model:  scientists could run thousands of "virtual experiments" overnight instead of one slow lab study at a time.

How the data will be collected

Volunteers will be scanned with two machines on campus.  A 7-Tesla MRI scanner, an unusually powerful magnet, shows in fine detail where in the brain activity happens.  A wearable OPM-MEG system, a helmet of sensors that picks up the brain's faint magnetic signals, shows when it happens, down to the millisecond.  People will watch, describe and recall thousands of videos while being scanned.  The models will be trained on Stanford's Marlowe computer cluster.

Imagine being able to test a new circuit intervention for depression by simulating its effects on a digital brain model before ever trying it in a patient

Researchers hope the models could eventually help match patients with depression to treatments that work for them, or spot early signs of Parkinson's or Alzheimer's disease.  None of that has been shown yet.

Lead funders Mark and Mary Stevens are longtime institute supporters; Mark Stevens is a venture capitalist and Nvidia board member.  Stanford said the data and models will be free for researchers worldwide.

Source:  Stanford Wu Tsai Neurosciences Institute announcement, Nov. 2, 2026.