A single human neuron is itself an extraordinarily sophisticated computing device
For generations, scientists have explained human intelligence by pointing outward — to the sheer number of neurons and the vast webs connecting them. Now, researchers at Hebrew University have found reason to look inward, discovering that individual human cortical neurons are themselves far more sophisticated computing devices than those of other mammals. The seat of what makes us cognitively distinct may reside not only in the architecture of the brain as a whole, but in the quiet, intricate labor of each cell.
- The long-held assumption that human intelligence is a product of scale — more neurons, more connections — has been meaningfully challenged by new cellular-level evidence.
- Using computational models and artificial neural networks as a kind of difficulty test, researchers found that human neurons were significantly harder to mimic than rat neurons, revealing a hidden layer of biological complexity.
- Human pyramidal neurons carry larger, more elaborately branching dendritic trees and NMDA receptors with sharper, nonlinear electrical responses — allowing a single cell to process information the way a small network would.
- The reversal of complexity patterns between cortical layers in humans versus rats points to an evolved advantage concentrated precisely where higher-order thinking happens.
- The findings are pushing both neuroscience and AI research toward a new design principle: sophistication at the unit level, not just scale at the system level.
For decades, the explanation for human intelligence has felt almost self-evident: we have an enormous number of neurons, wired together in staggeringly elaborate ways. A new study from Hebrew University's Edmond and Lily Safra Center for Brain Sciences, led by Idan Segev and Mickey London, quietly unsettles that story. The answer to what makes human cognition special, they suggest, may begin inside the individual cell.
The research team built detailed computational models of cortical pyramidal neurons — twelve from humans, twelve from rats — and then challenged artificial neural networks to predict each cell's behavior. The reasoning was elegant: the harder a neuron was to imitate, the more complex its underlying computation. By this measure, human neurons were dramatically more sophisticated, scoring an average of 0.3803 on the team's Functional Complexity Index compared to 0.2244 for rat neurons — a gap that was statistically consistent and far from trivial.
Two biological features appear to drive this difference. Human cortical neurons possess far larger and more elaborately branched dendritic trees, which can partition into semi-independent processing zones within a single cell. Meanwhile, human synapses — particularly those involving NMDA receptors — respond to incoming signals in a sharply nonlinear way, jumping in response around a threshold of 35 simultaneously activated synapses, where rat synapses respond in a far gentler, more linear fashion. Together, these properties allow a human neuron to function less like a simple switch and more like a miniature network in its own right.
Perhaps most telling is where this complexity concentrates. In rats, layer 5 neurons are the most computationally powerful. In humans, that distinction belongs to layer 2/3 — the very layer most expanded in the human cortex and most responsible for communication between cortical regions. Evolution, the findings imply, may have produced not just more neurons, but more capable ones, precisely where complex thought is coordinated.
The study carries acknowledged limits: incomplete experimental data left some active electrical properties unmodeled, and the Functional Complexity Index is relative rather than absolute. Yet its rankings held firm across different testing conditions. What remains open is whether still-unmeasured features of human neurons make the true gap even wider — and what it would mean for artificial intelligence if future systems were built not just larger, but genuinely more sophisticated at the level of each individual unit.
For decades, neuroscientists have pointed to an obvious fact to explain human intelligence: we have a lot of neurons, and they're wired together in staggeringly complex ways. But a team of researchers at Hebrew University has found something that complicates that tidy story. The answer to what makes human brains special may not live only in the sheer number of cells or their connections. It may live inside the cells themselves.
The study, led by Idan Segev and Mickey London at the Edmond and Lily Safra Center for Brain Sciences, along with collaborators from Vrije Universiteit Amsterdam, compared individual cortical neurons from humans and rats using a clever test. They built detailed computational models of these cells and then asked artificial neural networks to learn how each one behaves—to predict what it would do when given a particular input. The logic was straightforward: if an artificial system could easily mimic a neuron, that neuron was doing relatively simple work. If the artificial system struggled, missing spikes or getting the timing wrong, the biological cell was doing something more intricate.
The results were striking. Across 24 pyramidal neurons—12 from humans, 12 from rats—human cells scored far higher on what the researchers call a Functional Complexity Index. The average human neuron scored 0.3803 on this measure. Rat neurons averaged 0.2244. The difference was not marginal. It was statistically significant and consistent. Individual human cortical neurons, in other words, are far more sophisticated computing devices than their rat counterparts.
The source of this extra power appears to come from two places. First is shape. Human cortical pyramidal neurons have much larger dendritic trees—the branching structures that receive incoming signals—with longer arms and more elaborate branching patterns than those in rodents. When the researchers analyzed 58 different structural features of these cells, the strongest predictor of computational complexity was total dendritic area. The second strongest predictor was the combination of dendritic area with the longest bifurcation branch. What matters is not just that human dendrites are bigger, but how their size and branching pattern work together. Those sprawling trees can split into semi-independent regions, allowing different parts of the same neuron to process information somewhat separately. This compartmentalization appears to boost the cell's overall processing power.
But morphology tells only part of the story. The researchers also found differences in how human and rat synapses behave electrically. Human synapses, particularly those involving NMDA receptors, show stronger conductance and steeper voltage dependence than rat synapses. In the models, when researchers activated increasing numbers of synapses on a branch of a human neuron, the response jumped sharply and nonlinearly around 35 simultaneously activated synapses. Rat synapses under identical conditions showed a much gentler, more linear response. That nonlinearity matters because it means a human neuron can do more than simply add up its inputs. It can combine signals in richer, more layered ways depending on where and when they arrive—functioning less like a simple switch and more like a miniature neural network unto itself.
This finding reframes a long-standing debate about the origins of human intelligence. The conventional wisdom has held that our cognitive abilities arise mainly from scale: more neurons packed into a larger cortex, wired together in more elaborate patterns. The new work does not dispute that these factors matter. But it suggests that the internal sophistication of individual neurons may have played an equally important role in shaping what human brains can do. The pattern was especially striking when the researchers looked at different layers of the cortex. In rats, neurons in layer 5 were more complex than those in layer 2/3. In humans, the pattern reversed: layer 2/3 neurons were significantly more complex. This matters because layer 2/3 is expanded in the human cortex and handles much of the communication between different cortical regions. The implication is that human brains may have evolved not just more cells, but more powerful cells, particularly in the layers that enable complex thought.
The research does have limits. The models could not account for certain active electrical properties in human dendrites because the experimental data is still incomplete. The Functional Complexity Index itself depends on the specific artificial network used to test each neuron, so it is a relative measure rather than an absolute one. Yet the ranking of neuron complexity remained robust across different network architectures and input patterns. The findings open two practical doors. For neuroscience, they offer a framework for connecting the physical structure of neurons to their information-processing abilities. For artificial intelligence, they suggest that future systems might gain power not just by adding more units, but by making each unit more sophisticated and more biologically faithful. The deeper question—whether unmeasured features of human neurons make the gap even larger than this study could detect—remains open.
Citas Notables
People often think of a neuron as a simple switch that either turns on or off. What we show is that a single human neuron is itself an extraordinarily sophisticated computing device.— Idan Segev, Hebrew University of Jerusalem
La Conversación del Hearth Otra perspectiva de la historia
So the headline says human neurons are more powerful computers than rat neurons. What does that actually mean? Does one human neuron think?
No, not at all. A single neuron doesn't think. But it does process information in a more sophisticated way than we used to assume. Think of it like this: we've always known neurons are the building blocks of the brain. What this study shows is that the human building blocks are doing more work before signals ever spread across the larger network.
How did they measure that? You can't just ask a neuron how hard it's working.
They used artificial neural networks as a kind of test. They built detailed models of real neurons and then asked an AI system to learn how each one behaves. If the AI could easily predict what the neuron would do, that meant the neuron was relatively simple. If the AI kept getting it wrong, missing spikes or predicting them at the wrong time, that meant the neuron was doing something more complex.
And the human neurons were harder to predict?
Much harder. The human neurons scored nearly 70 percent higher on their complexity measure. But what's interesting is why. It's not just that human neurons are bigger. It's that they're shaped differently and they respond to electrical signals in a fundamentally different way.
Different how?
Human neurons have much more elaborate branching structures that receive incoming signals. Those branches can split the neuron into semi-independent regions, so different parts can process information at the same time. And when you activate multiple synapses at once, human neurons show a sharp nonlinear jump in response, while rat neurons respond more gradually. That nonlinearity means a human neuron can combine signals in richer ways.
So what does this mean for how we think about human intelligence?
It shifts the focus. We've always said humans are smart because we have more neurons and they're wired together more elaborately. And that's still true. But this suggests that part of the answer is that each individual neuron is already a more sophisticated processor. It's not just about the network. It's about what the building blocks themselves can do.