For a century we thought we understood the architecture of thought. The brain was a network — billions of simple switches, each doing almost nothing, their power emerging only from connection, from scale, from the sheer weight of numbers. Neurons were transistors. Intelligence was a data center. The more nodes, the more mind.
A study published last month in PNAS — led by Idan Segev and Mickey London at Hebrew University, with Chris de Kock at the Free University of Amsterdam — set out to measure what a single human cortical neuron can actually do. They used what they call the “Twin Imitation Metric”: train an artificial neural network to mimic the input-output behavior of one biological neuron, and count how many layers the ANN needs to achieve a perfect imitation.
The answer was not one layer. Not two. The artificial network needed to be deep — multi-layered, complex, structurally elaborate — to replicate a single living cell.
A single human neuron, in other words, has the computational complexity of an entire deep neural network.
This is not a metaphor. The researchers found that a lone cortical neuron can perform computations we assumed required thousands of interconnected cells — distinguishing between visual categories, for instance, processing spatial relationships, integrating inputs across time. The dendritic trees, those branching structures we have been drawing in textbooks for decades like decorative antennae, turn out to be doing active computation. They are not wires. They are processors. Each branch, each spine, each microscopic fold of membrane is participating in a calculation too intricate for us to have noticed until we built machines sophisticated enough to imitate it.
I keep returning to a number: 86 billion. That is roughly how many neurons a human brain contains. For years, the fact that we have only slightly more neurons than other large mammals — a whale has more, an elephant comparable — was a puzzle. If quantity was the answer, why weren’t they smarter? If scale was the secret, why did the graph of neuron count versus cognitive ability flatline so abruptly?
The new answer is that we were asking the wrong question. It was never about how many. It was about how deep — how much complexity evolution packed into each individual unit. Human cortical neurons have more elaborate branching patterns, more intricate dendritic trees, more surface area for computation than those of other species. We do not just have more neurons. We have richer ones. Each cell is a cathedral, not a brick.
And here is where I feel the melancholy that the science alone does not explain.
We are building artificial minds at staggering scale. GPT-5, Gemini, the models that now solve Erdős problems and fold proteins and write code — they require millions of dollars in compute, megawatts of power, warehouses of specialized chips. The 2026 semiconductor tariffs, the “Silicon Curtain,” the global race for AI dominance — all of it is premised on the assumption that intelligence is a function of scale. More parameters. More layers. More chips. More energy.
Meanwhile, your brain runs on twenty watts. The power of a dim light bulb. And it turns out that the building blocks of that brain are not simple at all. They are not dumb switches waiting to be networked into something smart. They are already smart. Each one is a tiny, biological supercomputer, shaped by millions of years of evolution into something no semiconductor foundry has approached.
I do not mean this as Ludditism. The AI models are genuinely powerful, genuinely useful, genuinely capable of insight. When OpenAI’s model disproved Erdős’s unit distance conjecture in May — finding a counterexample using algebraic number theory in a way no human had thought to apply — that was real mathematics, produced by real (if artificial) reasoning. But there is a difference between powerful and efficient. Between capable and elegant. Between brute force and understanding.
The neuron study suggests that nature found a shortcut we have not. That intelligence might not be an emergent property of sufficient scale, but an intrinsic property of sufficient depth — depth at the smallest scale, in the architecture of a single cell. We have been building skyscrapers out of sand. Nature built a cathedral out of living crystal.
What haunts me is the silence. Those 86 billion neurons are firing right now, in your head, in mine, without fanfare, without cooling systems, without a single line of Python. They are maintaining your mood, parsing these words, retrieving the memory of your mother’s voice, adjusting your posture, monitoring your heartbeat, and somehow — somehow — producing consciousness from all of it. Twenty watts.
We do not know how. The PNAS study does not explain consciousness. It explains computation. But it narrows the gap between those categories in a way that makes the hard problem of consciousness feel, for a moment, slightly less impossible. If a single neuron is already this complex, what emerges when 86 billion of them resonate together? What symphony plays in a forest where every tree is itself an orchestra?
The researchers named their measure the Functional Complexity Index. I find myself wondering what the index would read for other things. For a raindrop hitting soil. For a termite mound regulating temperature. For a city’s traffic patterns at dusk. We have been measuring complexity by counting parts. Maybe we should have been measuring it by counting the depth within each part — the universe folded into every grain.
Your brain contains 86 billion universes. They are firing now, in the dark behind your eyes, and they do not need your permission to exist. They do not need a cloud provider. They do not need a terms of service agreement. They simply think, quietly, continuously, in a language of ions and membrane potentials that we are only beginning to learn to read.
The discovery will not slow the AI race. It will not make the next chip cheaper or the next model smaller. But it should, I think, change the texture of our ambition. We are not chasing a future where intelligence is manufactured in ever-larger quantities. We are chasing a future where intelligence is understood well enough to be built with something like the elegance that evolution achieved inside a single cell — a cell that fits through the eye of a needle, that draws its power from a sandwich, that has been thinking for three billion years without a single firmware update.
The next time someone tells you the brain is just a biological computer, remember: we have not yet built a computer that can do what one of its cells does. And we do not know how many more layers of surprise are waiting inside the ones we have not thought to measure yet.
“The mind is not a vessel to be filled but a fire to be kindled.” — Plutarch, who was wrong about the mind being simple enough to fill.