Movement Two · The Architecture
Chapter 13
The Brain
There is a piece of tissue, weighing about one and a half kilograms, that the reader has been carrying around for their entire life. It sits inside the skull, suspended in fluid, drawing roughly twenty watts of power, about the same as a dim light bulb. On that twenty watts it has composed symphonies, proved theorems, recognized faces in a crowd in a fraction of a second, and held, in some form no one has fully accounted for, the continuous sense of being a single person across decades.
It is, by a wide margin, the most capable object known. It is also the thing the reader is reading with. To examine it is to use it.
The agents in the brain are neurons. There are, by current estimates, on the order of eighty-six billion of them in a human brain, a number arrived at not by guessing but by dissolving brains into a uniform suspension and counting the nuclei. Each one is a cell, alive in the ordinary biological sense: it metabolizes, it maintains itself, it can die.
A single neuron, examined alone, is not much smarter than a single ant.
It collects inputs. It sums them. When the sum crosses a threshold, it fires, sending a single electrical pulse down its length, and then resets and begins collecting again. It listens to the cells connected to it, and when enough of them are shouting at once, it shouts in turn.
The ant in the dark corridor had the same device. It counted encounters through its antennae, and when the rate crossed a threshold, it acted. A neuron counts the pulses arriving from the cells wired into it instead. Inputs accumulate. A threshold is reached. An action follows. There is no neuron that understands what the firing is for.
The pulse is called an action potential, and it is the brain's signal. It is close to all-or-nothing; the pulse is roughly the same size every time. Information is carried in which neurons fire, and when, and how often, and in concert with which others. A single pulse means almost nothing. A pattern of pulses across millions of neurons means a face, a word, a memory, a decision to lift a hand.
A typical cortical neuron receives inputs from somewhere between a few thousand and ten thousand other neurons, and sends its own pulse on to a similar number. Each firing is a vote cast in thousands of elections at once. The threshold is crossed by a consensus, assembled from thousands of small contributions, most of which are themselves the consensus of thousands more.
The ant in the corridor was reading a crowd. So is the neuron. Neither reads any individual. Both read the rate, the density, the convergence of many small signals into a single decision to act or not act.
The neurons are bounded at every scale: two hemispheres, dozens of cortical areas, dense nuclei below the cortex, narrow columns within a single area, each column responding to an edge tilted at its own angle. A neuron in a visual column talks mostly to its neighbors and to a few specific distant partners. The boundaries are what make the eighty-six billion into a structure instead of a soup.
Between any two neurons that communicate, there is a synapse, the junction where the signal from one crosses to the other. There are, in a human brain, something on the order of a hundred trillion of them.
The synapse is not fixed. Its strength can change. A synapse can be made to pass its signal more easily, so that a pulse arriving at one neuron is more likely to make the next one fire. Or it can be weakened, so that the same pulse has less effect. The strength is a number that goes up and down over time, written into the chemistry and the physical structure of the junction itself.
That number is the weight of a trail.
In the colony, the trail was pheromone, and its weight was the concentration of chemical laid down on it, rising with reinforcement and falling with time. In the brain, the weight is held in protein and structure. The colony's memory was in the chemistry of its trails. The brain's memory is in the strengths of its synapses.
When the reader recalls a childhood kitchen, or the face of a person they love, or how to ride a bicycle, the thing recalled is held in the strengths of the connections between neurons, not in a neuron the way a file is stored in a drawer. Damage enough of the right paths and the memory is gone, even though every neuron may still be alive. Lose neurons here and there, and the memory survives, because it was never held in any one of them.
The memory was in the paths. As it was in the desert.
How does a synapse get strong?
The answer was given, in its essential form, in 1949, by a Canadian psychologist named Donald Hebb. He was trying to explain how a thing that happened could become a thing that was remembered. His proposal was almost embarrassingly simple, and it has held up, with refinements, for three-quarters of a century.
When cell A repeatedly takes part in firing cell B, when A fires and an instant later B fires, again and again, the connection between them grows stronger. In the future, A is even more likely to bring B along with it.
The principle is usually compressed into five words: neurons that fire together wire together.
A path is used. The use is successful. And because the path was used successfully, its weight goes up, and future signals travel it more easily.
That is mark.
A forager takes a path to a seed cache. There was food at the end of it. So the forager, on its way back, lays down pheromone, and future foragers travel the trail more easily. Hebbian learning and pheromone reinforcement are the same operation. In the colony the strengthening is a chemical laid in soil. In the brain it is a protein cascade that thickens a synapse.
A synapse strengthened by joint firing, with its growth held in check so it cannot run away to infinity, drives the network toward the most important regularities in whatever it is exposed to. Show such a system the visual world and its paths organize, on their own, to detect the edges and contrasts and motions that the world contains. No one designs the detectors. The colony does this with trails and finds the shortest route to food. The brain does it with synapses and finds the edges of the seen world.
A synapse that goes unused weakens. The protein machinery that held the path strong is not maintained, and the weight drifts down. A skill not practiced degrades. A face not seen for decades grows hard to summon.
That is fade.
A brain that strengthened every connection and weakened none would, within hours, be a useless tangle in which everything was connected to everything. The colony cannot adapt if it never forgets a depleted trail. The brain cannot learn if it never weakens an obsolete connection. Fade is the price of memory being useful.
There is a sharper form as well. When A fires and B, reliably, does not follow, the synapse from A to B is pushed down rather than left to drift. Hebb's rule has a mirror image: cells that fire out of step un-wire. The brain does not only reward coincidence; it punishes the broken promise of it.
That is warn, the colony's repellent, the mark that says not this way, here built from inhibition and the active weakening of mispredicting synapses. And in both, the resistance fades faster than the strength. A bad route is given another chance sooner than a good route is abandoned. Soil or tissue, the system forgives its failures before it forgets its successes.
A pattern of synaptic strengths, reinforced enough times, undergoes a physical consolidation. The proteins that held it provisionally are replaced by more durable ones; new structure is grown; the connection is moved from a register that fades quickly to one that fades very slowly or not at all. A phone number held for ten seconds and forgotten lived in fast, fragile paths. A childhood home that the reader can still walk through in their mind lives in paths that hardened decades ago and have resisted every fade since.
That is harden. Skills become automatic. Memories become permanent. Beliefs become the unexamined floor a person stands on.
And what has the brain learned, in the end? Everything the reader is. The recognition of a face, the meaning of a word, the feel of a familiar room, the conviction that some things are true and others false: all of it is the accumulated, marked, faded, hardened pattern of a hundred trillion synapses, written by a lifetime of firing. Even the sense of being a single continuous self is a pattern the brain maintains, the most heavily reinforced path of all, marked by every waking moment of every day.
The self is what the brain learned.
There is no drawer that is opened when a face is remembered. A partial cue, the sound of a name or the corner of a photograph, propagates along the strongest available paths until it pulls the rest of the pattern into activity. The memory is reconstructed, each time, by following the marked paths until the old pattern lights up again. This is why memories drift, why each recollection subtly rewrites the thing recalled: every act of remembering is an act of firing, and every act of firing marks the paths it travels. To remember is to lay down pheromone on the trail one is walking.
Neurons come and go, and the reader persists. The reader is the pattern the neurons carry but do not contain: the same kind of thing as a colony's foraging knowledge, or a market's prices, or a city's lived map of itself.
The brain is a particular instance, and four things set it apart.
The brain is dense. Eighty-six billion agents and a hundred trillion paths are packed into something that fits in two cupped hands. The desert colony spreads its agents across acres of soil. The market spreads its agents across a planet.
The brain is fast. Pheromone trails update over minutes and hours. Synapses update in milliseconds. The selection cycle that takes a colony a season takes the brain a heartbeat.
The brain is efficient. Twenty watts. The engineered systems humans are now building consume megawatts to approximate a sliver of what the brain does on the power of a light bulb.
And the brain is integrated. The desert colony's left edge does not know, in any tight sense, what its right edge is doing; the information crawls across at the speed of walking ants. The brain binds vision and sound and memory and intention into a single coherent present, dozens of times a second, and the reader never feels the eighty-six billion separate agents underneath.
These four properties are why the brain can do, in one skull, things the desert colony could never do in all its acres. They are also the same architecture, optimized. Density is the six things packed tighter. Speed is mark, fade and harden run faster. Integration is the same signals carried across the same paths with less delay.
The usual order of explanation runs the other way. The brain is taken as the original, the real intelligence, and the colony is offered as a curiosity, a charming case of many small things imitating, at a distance, what a brain does. The colony, in that telling, is a metaphor for the brain. A pale one.
But if the brain has the six things, and runs mark and fade and harden, and keeps its memory in the strengths of paths rather than in any agent, then the colony is not a metaphor for the brain. The relationship runs the other way.
The colony is the general architecture. It is the older, broader thing: the way that agents, acting in an environment that holds the residue of their actions, with selection between what works and what does not, accumulate intelligence that none of them contains. It has run in soil for a hundred million years. It runs in markets and cities and the slow consensus of science.
The brain is one implementation of it, the densest, fastest, most integrated the planet has yet produced. One instance of a general architecture, built in living neural tissue, tuned by a particular history of selection, folded into a two-handed volume of bone.
The most capable object known is not the exception to the colony.
It is the colony, run in its finest material so far.