Science · Neuroscience

A Single Neuron Rivals a Deep Neural Network in Power

Neuroscientists found that one human cortical neuron can compute at the level of a small artificial neural network — challenging the long-held view of the neuron as a simple on-off switch.

The brain's basic unit, the neuron, has for decades been treated by computer models as a crude threshold device: sum up its inputs, fire if the sum is high enough, otherwise stay silent. A new analysis overturns that picture. Researchers at the Hebrew University of Jerusalem demonstrated that a single human cortical neuron possesses computational capabilities comparable to a small deep neural network.

A John Henry of the brain

In a simulated head-to-head, the team compared the computational power of an artificial neural network against a single biological brain cell, using advanced computer models and AI analysis. The result: the lone neuron matched — and in some tasks exceeded — a network of many artificial units. Where scientists expected a simple integrate-and-fire unit, they found a surprisingly sophisticated computation happening inside one cell.

Each human cortical neuron is a complex geometry of branching dendrites and a long axon. The dendrites receive thousands of synaptic inputs, and the cell integrates them in ways that go well beyond adding numbers. The finding points to a broader principle: the brain may compress intelligence into the computational richness of individual cells rather than relying solely on the number of connections between them.

The implication. If a single neuron already performs the kind of work engineers usually build a whole network to do, then the brain's 86 billion neurons may be far more computationally dense than current AI architectures assume. It also gives neuroscientists a framework for linking the physical shape of a brain cell to what it actually computes.

From a building block to a machine

The work matters for two reasons. First, it rewrites the textbook unit of neural computation: the neuron is not a dumb switch but a small processor in its own right. Second, it gives brain-computer interface research a clearer target. Understanding what a single cell computes makes it possible to read — and one day write — its activity with far higher precision.

It also reframes the comparison between biological and artificial intelligence. Modern deep learning stacks billions of simple units on top of each other to approach human-like tasks. The brain seems to achieve similar feats with far fewer "units," each one doing more of the heavy lifting internally. Closing that gap will likely require AI that imitates not just the wiring of neurons, but the computation hidden inside each one.

Knowledge takeaway: A single human cortical neuron computes at the level of a small deep neural network. Human neurons have ~86 billion of them, each a geometrically complex processor, not a simple on-off switch. The finding gives a framework linking a cell's physical shape to its computational role.