Movement One · The Colony
Chapter 1
Brain or Colony?
A nine-month-old child, sitting on a kitchen floor, watches a ball roll behind a chair. She crawls around the chair to the other side. She has never seen this ball, this chair, this floor before. No one has told her that objects continue to exist when she cannot see them. She arrives at it herself, in something under a second, using a brain not yet the size of an adult's and running on roughly the power of a dim lightbulb.
There is nothing in the desert that does this.
A harvester ant colony, for all its competence, cannot picture a seed that has rolled out of sight. It cannot hold an idea in its mind, because it has no mind to hold it in. It cannot reason about a thing that is not in front of it. It cannot be surprised. It will never, in a hundred million years, invent geometry.
The child will.
So before going any further, the honest question has to be asked. The prologue ended on an observation about ants — that the intelligence was not in any of them. But the most successful thinking machine anyone has ever encountered is not a colony. It is the thing reading this sentence. If intelligence were a contest, the brain would have already won it, and the rest of this book would be a long detour around a settled fact.
It is worth taking that possibility seriously. Seriously enough to make the strongest case for it that a careful person could make.
The case for the brain is this.
It is fast. A signal crosses the human cortex in milliseconds. A thought — a real thought, the kind that connects a smell to a memory of a kitchen forty years gone — assembles itself in less time than it takes to say the word thought. Nothing in chemistry moves that quickly. A pheromone trail takes minutes to build and minutes to fade. A brain runs at a speed the desert cannot approach.
It is integrated. Everything a brain knows is available to everything else it knows. The part that sees a face can talk instantly to the part that remembers a name and the part that feels affection and the part that moves the muscles into a smile. There is no shipping, no waiting, no relay. The whole of a brain's knowledge sits in one place, reachable at once. This is why a person can hear three notes of a song and feel, in the same instant, the year they first heard it, the room it played in, the person who was there. The colony has nothing like this. Information in a colony moves only as fast as an ant can carry it across the ground, and no ant ever holds more than the small piece in front of it. There is no place in a colony where everything comes together. In a brain, there is.
It reasons across domains. A brain that learned to throw a stone can use the same machinery to understand a thrown insult. A brain that learned to track a deer can track an argument. The abstractions transfer. A mathematician and a hunter and a poet are running the same organ, and the organ does not care which task it is given.
And it has produced everything. Language. Writing. The wheel. The proof that there is no largest prime number. The vaccine. The symphony. The very paper that described how to transfer value between strangers without a bank. Every artifact of civilization came out of brains, and only out of brains. No colony has written anything down. No colony has asked why the sky is dark at night.
A serious researcher, asked why artificial intelligence should be modeled on the brain, would point at all of this and say: because the brain is the only thing in the known universe that does what we are trying to build. We are trying to build a general reasoner. The brain is the only general reasoner there is. Why would anyone study insects?
It is a strong case. It is, on its own terms, almost airtight.
Now the other one.
The case for the colony does not begin with cleverness. It begins with the things a brain cannot do, no matter how clever it is.
A brain cannot scale. The human brain has about eighty-six billion neurons, and that number is fixed by the time a person is young. It cannot add more. A larger brain would not fit through the birth canal, would not cool itself, would not feed itself. There is a ceiling, and evolution hit it, and the species has been living under it ever since. Every human brain is approximately the same size as every other, and approximately the same size as a brain forty thousand years ago.
A colony has no ceiling. A particular species of ant might run a colony of a few dozen individuals or a few billion, and the rules do not change between the two. The same encounter-rate logic that allocates twelve foragers allocates twelve million. There is no architectural redesign at scale. The colony simply has more agents and the same substrate, and it works. A thing that runs unchanged from twelve to a billion is rare. The brain is not one of those things.
Damage tells the same story. Injure a small region of a brain and a person may lose the ability to recognize faces, or to form new memories, or to speak — failures specific and often permanent, because the knowledge was held in that tissue and nowhere else. Kill a tenth of a colony's foragers and younger ants take up foraging within hours; kill a third and the colony slows but persists, because no ant holds anything the colony depends on.
A brain forgets when it dies. Whatever a person learned across a lifetime — the route through a particular forest, the face of a particular friend, the trick for calming a particular child — leaves the world when the brain that held it stops. A library can preserve what was written down. But the written-down part is a sliver. The rest goes dark.
A colony's knowledge outlives every individual that holds it. The ants that forage a productive patch this year will all be dead within a year. The trail to the patch will not. It sits in the chemistry of the ground, reinforced by each ant that uses it, fading only if the patch stops paying. A colony observed for thirty years is composed, by the end, of entirely different ants than at the start — and it still knows where the good foraging is. The memory is not in the ants. It is in the substrate, and the substrate does not die when an ant does.
This is worth sitting with, because it inverts something that feels obvious. A person assumes that to remember a thing, something must hold the memory — a cell, a structure, a place in the head where the memory lives. The colony remembers without anything holding the memory in that sense. No ant carries the location of the patch. No ant could be dissected to find it. The knowledge is distributed across the chemistry of the ground and the behavior of thousands of ants and the rate at which they meet, and it exists only as long as the colony keeps performing it. Stop the colony and the memory is gone, not because it was erased, but because it was never stored anywhere that could be erased. It was a thing the substrate kept doing, not a thing the substrate kept.
There is also the matter of what the colony actually computes, which is easy to underrate because it is done so quietly. A colony with a hundred thousand foragers and a dozen food patches scattered across the desert is solving, every morning, a problem that a human operations researcher would recognize on sight: how to allocate a workforce across competing sites whose value keeps changing, with no site's value known in advance, and no time to stop and calculate. The colony solves it continuously, without a manager, without a map, and without any ant knowing there is a problem. When a patch runs dry, the trails to it fade and the foragers drift elsewhere within hours. When a rich new patch appears, a few scouts find it, their trail strengthens, and the workforce flows toward it. This is the same class of problem — finding good routes through a changing landscape of costs and rewards — that humans have spent serious effort teaching machines to handle. The colony has been handling it since before there were primates to watch.
And then there is the matter of behaving well.
A brain has to be taught not to do harm, and the teaching does not always take. Good behavior in a person is installed — by parents, by law, by conscience, by fear — on top of an organ that is perfectly capable of cruelty and frequently chooses it. The rules sit above the machine, and the machine can ignore them.
A colony's good behavior is not installed on top of anything. An ant that wastes the colony's resources, that forages a patch that does not pay, that follows a trail to nothing, simply stops being reinforced. Its path fades. The behavior dies out, not because anyone punished it, but because the substrate stopped feeding it. What does not work is not forbidden. It is starved. Bad behavior in a colony is not against the rules. It is against the physics.
The difference is worth being precise about, because it will matter later. A person who behaves well is a person whose impulse to behave badly has been overruled — by upbringing, by law, by the fear of being caught. The capacity for harm is intact; it has been suppressed. Remove the supervision, the law, the watching eyes, and the capacity is still there, ready. This is why human institutions spend so much of their effort on watching. The good behavior is a layer, and layers can be peeled back. The colony has no layer to peel. There is no point at which an ant is tempted to do harm and held back from it. The behaviors that would harm the colony are not held back. They are simply not reinforced, and so they fade, the way an unused trail fades, leaving nothing to peel back to. A colony is not a population of well-governed individuals. It is a structure in which the ungoverned outcome is already the good one.
A serious researcher, asked why artificial intelligence should be modeled on the colony, would point at all of this and say: because we are not trying to build one clever thing. We are trying to build something that scales without breaking, that survives the loss of its parts, that remembers past the death of any component, and that does the right thing because the structure makes the wrong thing unsustainable. The brain does none of these. The colony does all of them. Why would anyone study a single organ that cannot grow, cannot heal cleanly, cannot remember past its own death, and has to be bribed into decency?
That is also a strong case. It is also, on its own terms, almost airtight.
So there are two cases, each strong, each pointed at the other's weakness.
It is tempting to score them. Lay the dimensions side by side and add up the points.
On speed, the brain wins, and it is not close. Milliseconds against minutes. Whatever else the colony is, it is slow.
On scale, the colony wins, and it is not close. A fixed ceiling against no ceiling at all.
On integration, the brain wins. Everything reachable at once, against a colony where information moves only as fast as an ant can carry it.
On robustness, the colony wins. Graceful degradation against catastrophic, specific failure.
On memory, it depends on what is meant. A brain holds vastly more, vastly faster, for one lifetime. A colony holds less, slower, but across many lifetimes. One is deep and mortal. The other is shallow and immortal.
On alignment — on doing the right thing — the brain has to be governed and the colony governs itself. The brain's good behavior is a rule sitting above a machine that can break it. The colony's good behavior is the machine. One can be corrupted. The other has nothing to corrupt.
On generality — the prize, the thing the whole field of artificial intelligence is reaching for — the brain wins on range and the colony wins on uniformity. A brain can do almost anything once, brilliantly, within a life. A colony does a smaller set of things, but it does each of them with the same machinery, forever, and the machinery never needs redesigning.
There is one more line worth drawing, because it cuts across all the others. A human brain runs on about twenty watts — the power of a dim bulb — and it cannot be made to run on much more, because the skull would overheat. The brain bought its speed and its density at a fixed energy budget it can never exceed. A colony has no such budget. Add more ants and the colony draws more energy from the world, in seeds and sun, with no ceiling and no overheating. The brain is a marvel of efficiency precisely because it had no choice. The colony was never forced to be efficient, so it was free to be large.
Add it up and the score is a tie that refuses to resolve. Each architecture wins exactly where the other loses, and loses exactly where the other wins. That is not the look of a fair fight between two contestants. That is the look of a trade-off — the same currency spent two different ways.
Which is the first hint that the question is wrong.
If the brain and the colony were two different kinds of thing, one of them ought to be simply better, the way a calculator is simply better than an abacus at arithmetic. Instead they are mirror images. Where one is fast it is fragile. Where the other is robust it is slow. The strengths and the weaknesses come in matched pairs, traded against each other along a single axis. Two things that trade along the same axis are usually not two different things. They are one thing, tuned two different ways.
So it is worth doing what a careful person does when a comparison keeps coming out even. Stop comparing the outsides. Look at what is actually happening inside each one.
Open a brain and look closely — not at the whole organ, but at what it is made of.
It is made of cells. About eighty-six billion of them in the part that does the thinking, each a small, simple unit. A single neuron is not clever. It receives signals from its neighbors. It adds them up. When the sum crosses a threshold, it fires a signal of its own to the cells it connects to. When the sum does not, it stays quiet. Receive, sum, compare to a threshold, act. That is most of what a neuron does.
A single neuron, removed from the brain and kept alive in a dish, is one of the least impressive things in biology. It will fire. It will fall silent. It has no goals, no understanding of its situation, no capacity to reason about what to do next. In isolation, it is almost not intelligent at all.
This should sound familiar.
It is, very nearly word for word, what the prologue said about a single ant.
There is something else the count makes plain. A neuron is slow. It can fire, at most, a few hundred times a second — millions of times slower than the switches in a pocket calculator. No single neuron could do arithmetic a child finds easy. Whatever a brain does that is impressive, it does not do because any of its parts is fast or clever. It does it because there are so many parts, acting at once, and because of how they are arranged. The cleverness is not in the unit. It is in the arrangement.
The neuron does not act alone any more than the ant does. It acts in an environment — a dense thicket of connections to other neurons, each connection a junction across which signals pass. A single neuron may share these junctions with thousands of others, and the thicket as a whole holds them by the hundreds of trillions. And those junctions are not fixed. When two neurons fire together, often, the junction between them strengthens. Signals cross it more easily afterward. When two neurons stop firing together, the junction between them weakens, and signals cross it less easily, until it fades. A connection that carries useful traffic is reinforced by the traffic. A connection that carries nothing fades.
The colony deposits a trail on the ground, strengthens it by use, lets it fade when the use stops. The brain deposits a connection between two cells, strengthens it by use, lets it fade when the use stops.
The ground is one substrate. The thicket of connections is another.
The marks are chemical in the desert and they are also chemical in the skull — the brain strengthens and weakens its junctions through chemistry too, a different chemistry, doing the same job. In both cases, simple units act. In both cases, what they do leaves a mark on the thing between them. In both cases, the mark fades unless it is reinforced. In both cases, the next unit's action is shaped by the marks the previous ones left.
Consider what this means for a single memory. A person learns, once, that a particular stretch of pavement is icy in winter. The lesson is not written down in any cell. It is held in a set of connections that grew slightly stronger the day they slipped — the junctions between the neurons that fired together when the foot went out from under them. Walk that pavement again and those strengthened connections fire more readily, and caution arrives before the conscious thought does. Years pass. The neurons that held the lesson may themselves change, lose and grow connections, age. The memory persists anyway, because it was never the property of those particular cells. It was the property of the pattern of connection between them, reinforced by use, ready to fade if never used again. The lesson is held in the substrate, in the trained weave of junctions that carries the marks of everything the person has done before — exactly as the location of a food patch is held in the trained weave of trails that carries the marks of everything the colony has done before.
The reasoning a child does, watching a ball roll behind a chair, is not done by any neuron. No neuron knows about balls or chairs or the persistence of objects, just as no ant knows about foraging strategy or territorial defense. The neuron knows only its inputs and its threshold. The understanding is not in the neurons.
It is in the substrate. In the trained thicket of connections that holds the marks of everything the child has done before. The neurons are doing the work. The substrate is doing the thinking.
This was the sentence the prologue used for the colony. It turns out to describe the brain just as exactly.
So the question that opened the chapter — brain or colony, which is the right model for intelligence — was the wrong question, the way earth or sky, which is the right place for weather is the wrong question. The brain is not an alternative to the colony architecture. The brain is the colony architecture, run in a different material, tuned for speed at the cost of scale, packed into a space the size of two fists, and wired so densely that the agents can touch in milliseconds instead of minutes.
The colony spreads its agents across acres of desert and connects them with chemistry that drifts on the air. The brain packs its agents into a skull and connects them with chemistry that leaps across a gap a few millionths of a metre wide. One is sprawling and slow and immortal. The other is dense and fast and mortal. But the move is the same move. Simple agents, acting on what they sense. A substrate between them that holds the marks of what they did. Marks that strengthen with use and fade without it. Behavior that survives because it was reinforced, and dies because it was not.
This is why the scorecard came out even. The two architectures were never competing. They were the same design, spent two different ways along a single budget — the budget of how to arrange simple agents and the substrate between them. Pack the agents close and connect them fast, and you get speed and integration, at the cost of a hard ceiling on how many you can have and how long they can last. Spread the agents wide and connect them slowly, and you get scale and robustness and a memory that outlives any of them, at the cost of speed. Every point the brain won and every point the colony won was the same trade, made in opposite directions. Speed against scale. Density against robustness. A deep mortal memory against a shallow immortal one. There was no contest because there was only ever one architecture, tuned to two settings.
And once that is seen, the matched strengths stop looking like an accident. They look like what you would expect if both things were the same kind of system, built from the same parts, obeying the same constraint that whatever you gain on one axis you pay for on another. Two genuinely different designs would not trade so cleanly. These trade cleanly because underneath they are not two designs.
The child on the kitchen floor and the colony in the New Mexico dust are not two answers to the question of where intelligence comes from.
They are the same answer, written twice.
Which means the interesting question was never which architecture wins. Both architectures are the same architecture. The interesting question is the one left standing once that is seen.
What is the architecture that is running in both?