Stanford Aims to Build a Digital Twin of the Human Brain

Stanford University's Wu Tsai Neurosciences Institute has launched an initiative to build a "digital twin" of the human brain, training an AI foundation model on approximately 14,000 hours of brain-imaging data collected from roughly 100 people over the next five years. Billionaire venture capitalist Mark Stevens and his wife, Mary Stevens, are funding the Stanford Digital Brain Project, with additional support from philanthropist Clara Wu Tsai; the institute declined to disclose the funding amount.

How will the data be collected?

The first of 10 planned data-collection "campaigns" will scan 10 participants about twice per week for a year, using two complementary modalities. The first is 7-tesla functional MRI (7T fMRI), which tracks blood-flow changes at roughly millimeter spatial resolution but lags neural activity by several seconds. The second is optically pumped magnetometer magnetoencephalography (OPM-MEG), a technique that detects the brain's magnetic fields on a millisecond timescale but localizes their sources less precisely.

Participants will initially watch videos and perform simple decision-making tasks during their scans. Eventually, they will play video games and complete tasks in virtual reality.

What is the scientific bet?

The wager, according to Dan Yammins, associate professor of psychology and computer science at Stanford and a faculty scholar at the Wu Tsai Neurosciences Institute involved in the initiative, is that AI models trained on large datasets will accurately predict brain activity and can generate hypotheses for later testing in people. Rather than training on internet-scale data as large language models do, the project will build on existing AI models that already predict some brain activity, he says.

Mariya Toneva, a faculty member at the Max Planck Institute for Software Systems who is not involved in the project, said the project "is very needed, and I'm happy that somebody was able to commit the resources and the time to do this." Even supporters acknowledge that no one can say how much data will be enough, and a model that predicts the brain well may still reveal little about how it works.

How does the Meta project compare?

The Stanford effort mirrors a parallel program at Meta, also called the Digital Brain Project, launched in August. Nikolaus Kriegeskorte, professor of psychology and neuroscience at Columbia University, is not involved in the Stanford initiative but is part of the Meta project, which he says shares "the same general vision" — and the same name. Meta donated $15 million to its project, which aims to collect 15,000 hours of imaging data. Nine research teams have received funding so far and initially plan to image at least 90 participants performing various tasks, including interactive games.

Whether either project can gather enough data to predict how the human brain computes remains, in Kriegeskorte's words, "a very difficult question that I don't have a good answer to." Current AI-based models mimic only portions of the brain, such as the visual cortex, and predict only about half of the variance in responses observed in those regions; modeling the entire brain will likely require far larger datasets than currently exist, he notes.

What are the skeptics saying?

Some researchers question whether foundation models — large AI models trained on broad data — will transfer to neuroscience as successfully as they have to language. "I'm kind of skeptical of all ventures in science and engineering where people are like, 'Name hard problem X. Let's build a foundation model for problem X,'" said Bing Wen Brunton, professor of biology at the University of Washington, who is not involved in the project.

Brunton expressed "a lot of respect" for the Stanford researchers and their past work in the field. But no one yet knows how brain models scale with data, she cautioned. "What's the evidence that this is going to work, besides that we're hoping it does?"

The first scanning campaign's dual-modality dataset — millimeter-resolution hemodynamics paired with millisecond-resolution magnetic fields — will offer an early test of whether densely sampled individuals can anchor a predictive whole-brain model.