Engineers are building a car. Some assemble an exact replica of the engine, every piston and spark plug, but the car doesn't move: there's no steering wheel or wheels. Others build a fast self-driving go-kart, but its engine is made of playdough and doesn't resemble a real one. This is roughly the situation in brain science. One approach—'whole-brain modeling'—recreates neural networks with terrifying detail, down to individual cells. Another—'neuroconnectionism'—teaches artificial neural networks to play Go and translate texts, but their internal workings only vaguely resemble a living brain.

The authors of the new work, including Ashish Vaswani, propose to cross these approaches. They call the result a 'functional whole-brain model.' It's like adding real traffic flows, jams, and traffic lights (dynamics) to a detailed 3D map of all city streets (structure), and crucially, the ability for residents to get to work and back home (function). The model must meet four conditions: rely on real connectivity maps between brain regions, operate in continuous time like real neurons, solve practical tasks, from face recognition to memory, and produce signals comparable to scanner data.

The authors sketch a roadmap for years ahead. The immediate goal is to build a simple working prototype. In the long run, such models could let us peer into the very essence of thought and perhaps learn to fix its malfunctions. After all, if you have an accurate simulation of a city, you can not only predict a traffic jam but also understand how to clear it.