ONE Brain or Colony?

Movement One · The Colony

Chapter 1

Brain or Colony?

9 min read · 2,050 words


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 will never, in a hundred million years, invent geometry.

The child will.

The most successful thinking machine anyone has ever encountered is the thing reading this sentence. If intelligence were a contest, the brain would have already won it.

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. A pheromone trail takes minutes to build and minutes to fade.

It is integrated. Everything a brain knows is available to everything else it knows. 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. Information in a colony moves only as fast as an ant can carry it across the ground. 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 track a deer can track an argument. A mathematician and a hunter and a poet are running the same organ.

And it has produced everything. Language. Writing. The proof that there is no largest prime number. The symphony. Every artifact of civilization came out of brains. 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: we are trying to build a general reasoner, and 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 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. A larger brain would not fit through the birth canal, would not cool itself. There is a ceiling, and evolution hit it, and the species has been living under it ever since.

A colony has no ceiling. The same encounter-rate logic that allocates twelve foragers allocates twelve million. The colony has more agents and the same ground beneath them, and it works.

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 — leaves the world when the brain that held it stops.

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. By the end of its life a colony is made of entirely different ants than at the start, and it still knows where the good foraging is.

No ant carries the location of the patch. No ant could be dissected to find it. Stop the colony and the memory is gone. It was never stored anywhere that could be erased.

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. 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 forages a patch that does not pay, that follows a trail to nothing, stops being reinforced. Its path fades. What does not work is starved.

A person who behaves well is a person whose impulse to behave badly has been overruled. The capacity for harm is intact. Remove the supervision and the capacity is still there. 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. The behaviors that would harm it fade, the way an unused trail fades. In a colony, 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: we are trying to build something that scales without breaking, that survives the loss of its parts, 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.

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.

On speed, the brain wins, and it is not close. Milliseconds against minutes.

On scale, the colony wins, and it is not close. A fixed ceiling against no ceiling at all.

On integration, the brain wins. On robustness, the colony wins.

On memory, 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 long-lived.

On alignment, the brain has to be governed and the colony governs itself.

On generality — 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, and the machinery never needs redesigning.

Add it up and the score is a tie that refuses to resolve. Each architecture wins where the other loses. 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 better, the way a calculator is 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.

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 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.

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. In isolation, it is almost not intelligent at all.

This should sound familiar.

The neuron does not act alone any more than the ant does. It acts in a dense thicket of connections to other neurons, each connection a junction across which signals pass. And those junctions are not fixed. When two neurons fire together, often, the junction between them strengthens. When they stop firing together, it weakens, until it 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.

A person learns, once, that a particular stretch of pavement is icy in winter. The lesson is held in a set of connections that grew slightly stronger the day they slipped. 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 change, lose and grow connections, age. The memory persists anyway, because it belongs to the pattern of connection between them, reinforced by use, ready to fade if never used again — as the location of a food patch is held in the trails that carry 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, just as no ant knows about foraging strategy. The neuron knows only its inputs and its threshold. The understanding is held in the trained thicket of connections that carries the marks of everything the child has done before. The neurons are doing the work. The substrate is doing the thinking.

So brain or colony 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, 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. 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.

This is why the scorecard came out even. 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. There was only ever one architecture, tuned to two settings.

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.

The interesting question is the one left standing once that is seen.

What is the architecture that is running in both?

2 of 25

100 Million Years Ahead

Prologue

  1. One Ant, August 1993

Movement One · The Colony

  1. 1Brain or Colony?
  2. 2What the Ants Are Doing
  3. 3How the Ant Decides
  4. 4The Pheromone Trail
  5. 5The Castes
  6. 6How a Colony Survives a Decade
  7. 7The Queen Is Not in Charge

Movement Two · The Architecture

  1. 8The Six Things Every Colony Has
  2. 9The City
  3. 10The Market
  4. 11The Scientific Community
  5. 12The Body
  6. 13The Brain
  7. 14The Language
  8. 15The Ledger
  9. 16Why the Pattern Holds

The Hinge

  1. 17The Two Materials

Movement Three · The Implications

  1. 18What Superintelligence Actually Is
  2. 19The Ceiling of the Single Model
  3. 20Alignment Is a Substrate Property
  4. 21What Civilization Already Is
  5. 22The Next Hundred Million Years

Epilogue

  1. A Note on Reading

Apparatus

  1. Notes on Sources