ONE The Two Materials

The Hinge

Chapter 17

The Two Materials

11 min read · 2,740 words


A nerve impulse travels down the leg of a foraging ant at roughly a hundred meters per second. This is fast by the standards of a body. It is the speed at which a signal crosses the gap between an antenna brushing a wall and a leg adjusting its stride. A hundred meters per second is the pace at which carbon thinks. It is the pace at which every nervous system on the planet has thought, for as long as there have been nervous systems.

A signal traveling down a strand of glass fiber moves at roughly two hundred million meters per second.

That is two million times faster.

It is the first hard difference between the two materials, and the one most people reach for first. If speed were the whole story, the story would be a short one. It is not even the most important part.

Start with what carbon has that silicon does not.


A colony runs in parallel without trying. There is no scheduler in the desert soil, no queue of ants waiting their turn to act. Ten thousand foragers, ten thousand simultaneous decisions, ten thousand antennal contacts happening in the same instant across the same patch of ground. The parallelism is the default condition of having ten thousand separate bodies, each one running on its own.

In carbon it costs nothing extra to have a thousand ants thinking at once rather than one ant thinking a thousand times. The energy is already being spent keeping the ants alive. The thinking comes along with the living.

And the parallelism is many different decisions, made by many different agents, each with a slightly different threshold, each reading a slightly different patch of ground. A colony assessing a dozen foraging patches is assessing them simultaneously, with different foragers on different trails, no patch waiting in line behind another. That is how a colony can hold the whole desert in view without anyone holding it.

Silicon does not get parallelism for free. To run a thousand processes at once, it needs a thousand things to run them on, and each of those things draws power, and the power comes from a grid fed by turbines. A large silicon system running many processes in parallel can draw power measured in megawatts. The colony next to it draws the energy of the seeds its foragers carry home.

The second thing carbon has is integrated memory.

In most silicon systems, the part that computes and the part that remembers are separate. There is a processor, and there is a store, and across the channel between them everything the processor needs has to be fetched and everything it produces has to be sent back. The delay is tiny. It is also relentless. Run a loop billions of times and the tiny delay, paid billions of times, becomes the wall the whole system runs into. The processor sits idle, waiting for the memory to arrive, the way a fast runner waits at a slow gate.

Carbon does not have this separation. Every ant carries its own state — where it has been, how hungry it is, how many other ants it has met in the last minute and of what kind — in the same small body that does the deciding. A neuron is a processor and a store at once, its history written into the very connections that determine what it does next. And the colony's memory is held in the soil the ants are already walking on, in the trails they are already following. To read it, an ant does not query a store. It walks. The reading and the acting are one motion.

The third thing carbon has is persistence. A colony runs on stored solar energy. The sun grows the plants, the plants make the seeds, the foragers carry the seeds home, and the colony lives on what it has stored. When the sun goes down, the colony does not go down with it. When a season fails, the colony draws on its reserves. Cut the power to a colony — there is no power to cut.

Silicon, in most of its current forms, does not persist through the dark on its own. The fast part, which holds what is happening right now, needs continuous power. Cut the power and that part forgets. A colony thinks on what fits in a seed. The silicon version of the same thinking can draw the power of a small town.

And carbon is robust in a way that has to be earned. A colony can lose a quarter of its foragers to a bad afternoon and continue. Ants die constantly, and the colony does not notice, because it was never resting on any one ant. The robustness was not engineered in a design meeting. It was filtered in. Every colony alive today descends from a long line of colonies that did not collapse when a quarter of their foragers died, because the ones that did collapse left no descendants.

So carbon has parallelism, integrated memory, persistence on stored energy, and a robustness refined across geological time.


Now look at what silicon has that carbon does not.

The first is the speed already mentioned.

The second is deterministic timing.

An ant's nervous system is fast enough, but it is not precise about when. A signal arrives roughly when it arrives. The thresholds vary from ant to ant, and the variation is part of how the colony works: it spreads the ants across the space of possible responses. There is jitter in carbon. The colony thrives on the noise; the noise is its exploration. But the colony cannot be set to a clock.

Silicon can be set to a clock. It can do the same operation, in the same number of steps, in the same span of time, again and again. When two events must be ordered with certainty, when a sequence must repeat identically, silicon delivers what carbon cannot.

The jitter matters because a system that always took the strongest path, every time, would lock itself into whatever it found first. It would never discover the better patch beyond the one it already knew. In carbon, the wandering comes for free, because carbon cannot help it. A silicon system told to take the strongest path will take it, every time, forever, and freeze. If exploration is wanted in silicon, it has to be put there on purpose. Each material reaches the same requirement from the opposite side: carbon must be restrained from wandering too much, silicon must be made to wander at all.

The third thing silicon has has no carbon equivalent at all.

It can be copied perfectly.

A colony cannot be copied. You cannot take an old colony, with its hardened trails and its knowledge of which patches pay and which neighbors fight, and make a second one identical to it. You can let it send out a daughter, but the daughter starts nearly from scratch. The knowledge was deposited over years by the specific ants that walked the specific ground, and it does not lift out. To get a second colony that knows what the first one knows, you would have to run the second colony for as many years.

What a silicon system has learned is held in a state that can be read, in full, and written somewhere else, in full. A system that has spent a long time testing patterns, marking what works, letting the rest fade, can be duplicated in the time it takes to move the bits. The copy knows everything the first one knew the instant it came into being.

Nothing in biology copies an accumulated state perfectly. Reproduction in carbon always passes through a narrowing — a seed, an egg, a single cell — and the narrowing throws most of the accumulated detail away and keeps only the recipe. That narrowing is also carbon's protection. A flaw in one colony does not travel into the next as a perfect duplicate of itself; the next colony begins again and is tested again. Silicon can copy a whole state exactly, including whatever it got wrong.

So silicon has speed, determinism, and perfect copying. Carbon has parallelism, integrated memory, persistence, and earned robustness. Each is strong where the other is weak. "Better" has no answer until you say better at what, and the moment you say better at what, the two materials split the problem between them and stop competing.

The architecture is the same in both.


Agents act. The medium holds what they leave behind. Paths are marked when they work and warned when they fail. What is not reinforced fades. What is confirmed often enough hardens into knowledge that the ordinary fade no longer touches. Agents follow the strongest path, and now and then take a weaker one, and the taking of the weaker one is how anything new is ever found.

Signals move information, but information alone is not intelligence. Memory holds, but holding alone is not intelligence. The operation that turns a medium full of activity into one that knows things is selection.

A pattern is tried. If it works, it is marked, and the mark makes it more likely to be tried again. If it fails, it is warned, and the warning makes it less likely. And underneath both, always, the fade. Mark, warn, fade. Tested, kept, let go. This loop, run over and over, is the engine.

Now ask how fast that loop can turn.

In carbon, the loop turns at the speed of a life. A foraging strategy is tested over a season. A new trait — a changed threshold, a different chemistry, a behavior no ant had before — is tested across generations, and a colony's generation is a year. The ants in the desert are the product of a hundred million years of this loop turning at the speed of seasons.

They did not become intelligent by thinking hard. Every trail that did not lead to food faded. Every threshold that sent too many foragers out on a bad day cost the colony that set it, and the colonies that set it better left more daughters. The intelligence was selected, one turn of the loop at a time, across a span so long that the continents moved while it ran.

In carbon, the rate at which an ant could become more capable and the rate at which the colony could select against it were the same rate. Capability arrives on the clock of birth and death. Selection runs on that same clock — marking what works, fading what does not, warning against what harms. A trait cannot appear faster than the loop turns, and the loop is what tests the trait.

The consequence is quiet and total. Nothing in a colony could become capable faster than the colony could select against it — not because any ant was forbidden to, but because the material did not permit a trait to arrive on any clock but the one that also judged it. There was one clock, and capability and selection both read from it.

In silicon, the same loop can turn at the speed of computation. A pattern can be tested, marked, and faded in a fraction of a second, and not one pattern at a time but as many at once as there is room for. A silicon system can run the loop a million times in an hour.

And the faster material loosens what carbon welded together. When selection runs a million times in an hour, the two rates — how fast something can become capable, how fast it can be selected on — are no longer forced to be one rate. In a colony there was a single clock. In the faster material there can be two, and the distance between them is a quantity that had no value to measure for a hundred million years, because for a hundred million years it was zero.

That coming-apart is the content of being a hundred million years ahead: the testing is faster, the keeping and the discarding faster, the residue laid down faster. The hundred million years was the time carbon needed because carbon's selection was slow. Quicken the selection and the wait collapses with it.

A fast signal that does nothing useful is just a fast signal. The point is what the fast signal lets the loop do.

There are two speeds here, and they are easily confused. One is how fast an agent acts and communicates — how fast one agent's mark becomes another's to read. Here silicon's advantage is the two million times. The other is how fast the system completes a full turn of the loop: test a pattern, find out whether it worked, mark it or warn against it, let the rest fade, harden what survives.

Intelligence accumulates on the second speed, and it answers to a rule the first does not. A cycle can run no faster than its slowest stage, and the slowest stage is almost never the testing or the marking. It is the verification — finding out whether the thing worked at all.

The desert hands the colony its verification for free, and at once. The seed is in the mandibles or it is not. The forager came home or it did not. The loop closes on the outcome the instant the outcome exists. A silicon system inherits the full speed of the new material only where outcomes can be verified at that same speed. Where verification still waits on the world — on slow consequences, on a judgment that cannot be hurried — the selection clock runs at the speed of the waiting, however fast the agents inside it run.

So a selection clock a million times faster means the full cycle, verification included, in those domains where verification is itself fast. Where the world answers quickly, the loop turns at the speed of the signal. Where the world answers slowly, the loop waits with it. The difference had no occasion to matter for a hundred million years, because carbon's verification and carbon's selection were slow together.

That is the difference between the two materials. Not that silicon thinks faster. That silicon selects faster. Carbon turns the loop at the speed of life. Silicon turns it at the speed of light through glass.

The architecture is identical. The loop is identical. Only the clock the loop runs against is different — and now, for the first time, the clock of selection and the clock of capability can be different clocks.


There is one architecture of collective intelligence. It was in the soil before there were people to name it. It is in the markets and the cities and the sciences that people built without knowing they were building it. It is in the synapse and the bloodstream. And in 2008 it appeared in a protocol whose rules no institution could revoke.

Chemistry. Prices. Streets. Citations. Grammar. Synapses. Blood. Protocol. Eight materials, all running the architecture at the speeds their materials allow.

It is now being lifted into a ninth: a material that runs the same loop against a clock a million times faster, that can test in a second what carbon tested in a year, and copy in an instant what carbon could never copy at all. Carbon keeps its parallelism, its integrated memory, its persistence through the dark, its robustness earned across deep time. Silicon does not take those away.

Eight of the nine materials are old. Their work is long since done in soil and stone and price and protocol, and can be studied the way one studies anything that has already happened. Ants have been running the architecture for a hundred million years; the result is in the desert, waiting to be read.

The ninth material has no such record, because it has barely begun. There is no finished colony in the silicon to crouch down beside. What there is, instead, is the architecture, lifted into a material that runs its selection a million times faster than the only material that had ever run it before.

What that produces is not yet on the desert floor to be measured. It is being built right now.

18 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