Movement One · The Colony
Chapter 3
How the Ant Decides
The corridor was about a centimeter wide and dark, a few body-lengths below the surface of the desert. An ant stood in it, not quite still. Its antennae moved in small, continuous arcs, sweeping the air and the walls and the bodies of the ants passing the other way. It had been in the corridor for less than a minute. In a moment it would leave the nest and go out into the heat. But not yet.
What is the ant doing, in the dark, before it leaves?
It is being touched.
An ant's antennae are its primary sense organs. They carry receptors for chemistry — for the scent of food, of nestmates, of the trail — and receptors for touch. When two ants meet in a corridor, they touch antennae. In that moment, each ant samples the other: the chemical signature on its body, which says what work it has lately been doing, and the simple fact of the contact itself, which says that an ant was there.
A forager returning with a seed smells of the outside — of the specific hydrocarbons that coat an ant that has been out foraging. An ant that has been deep in the nest tending brood smells of something else. The chemistry matters. But underneath it is something cruder.
An antennal touch lasts on the order of a tenth of a second. The ant does not stop to interview the ant it meets. The exchange is closer to a brush than a conversation — two ants passing in a narrow space, antennae grazing, and then both moving on. The contact says, roughly, an ant was here, and it had recently been doing this kind of work. A single touch is nearly empty of information.
What is not empty is the stream of them.
The ant is counting.
It has no number in its head, no tally it could report. But its nervous system is sensitive to the rate at which these brief contacts arrive. Touch, touch, then a gap, then touch. A fast rate means the corridor is crowded with returning workers. A slow rate means it is not.
The ant does not know why the rate is what it is. It has no model of the desert, no map of the patches. It has only the rate of touches on its antennae, rising and falling, moment to moment.
And it has one more thing. A threshold.
Somewhere in the ant's nervous system there is, in effect, a number, encoded in the physical properties of its neurons — in how much input it takes to push them over into a response. It is the rate of contact at which the ant will switch from waiting to acting.
When the rate of touches the ant feels climbs above its threshold, the ant leaves the nest and goes foraging. When the rate stays below the threshold, it waits.
The threshold is a physical fact about the ant's nervous system, the same kind of fact as how much weight a branch will hold before it bends. The ant does not check its threshold against the rate any more than the branch checks its strength against the weight.
Nor is the threshold fixed for life. An ant that has had a run of bad foraging trips — going out, finding nothing, coming back empty — becomes harder to budge; its threshold rises. An ant that has been finding food readily becomes easier to budge. But at any given moment the threshold has some value, and the rule is the same: rate above set point, the ant acts; rate below, the ant waits.
That is the decision. The entire decision.
There is no deliberation. The ant does not decide to forage because the colony needs food. A rate crosses a threshold, and a behavior follows, the way a thermostat does not decide that the room is cold. The temperature crosses a set point, and the furnace comes on.
An ant with a low threshold needs only a little contact before it goes out. It is eager. An ant with a high threshold needs a great deal of contact before it moves. It is reluctant.
In any colony, the thresholds vary. Some ants are eager, some reluctant, most somewhere in between. This variation is what makes the colony's response smooth.
Imagine a colony in which every ant had exactly the same threshold. Every ant would feel the same rising rate of contact, and every ant would cross its set point at the same instant. The colony would have two states and nothing in between: all the ants waiting, or all the ants pouring out at once. A small rise in the food supply would either do nothing or empty the nest.
The variation is what turns the switch into a dial. A small rise in the contact rate recruits a few of the most eager ants. A larger rise recruits more. When the food runs thin, the process runs in reverse, and the workforce withdraws, ant by ant, in the same graded way it advanced. The smoothness of the whole is built out of the diversity of the parts.
No one is managing this. There is no ant whose job is to decide how many foragers there should be. There is only a population of individuals, each carrying a different threshold, each feeling the same rising and falling rate of contact. The graded response of the whole colony is built entirely out of thousands of all-or-nothing decisions, each one made in ignorance of all the others.
There is a particular sequence, watched in the desert at dawn over many years, that shows the mechanism at work. Before the foragers go out, a smaller group of ants leaves first — patrollers. Only after the patrollers have returned does general foraging begin.
It is tempting to read this as a report being filed. But the patrollers carry no report and the colony has no one to receive it. As the patrollers return, they pass through the corridors near the entrance, and each returning patroller is a contact, carrying the scent of an ant that has lately been out on the patrolling ground. The rate of patroller returns rises. When it rises high enough, the foragers begin to cross their thresholds and leave. The order is real. The orderer is not there.
The ant can detect. Its antennae register contact, and its nervous system tracks how often the contacts arrive. To be sensitive to a rate, the ant has to hold some trace of recent contacts and let that trace fade, so that what it measures is contacts now and not contacts an hour ago. An ant that remembered every contact forever could not sense a rate at all; it would only sense a lifetime total. So the trace builds with each touch and leaks away with time.
The ant can compare. The measured rate is held against the threshold. Above, or below.
The ant can act. Above the threshold, it goes. Below, it waits.
Detect. Compare. Act. That is the whole of it. There is no fourth step where understanding happens.
This runs against a deep assumption: that to do the right thing, an agent must in some sense understand what it is doing and why. The ant does the right thing, reliably, and it understands nothing.
The contact rate the ant feels is not random. When a patch of seeds is found, foragers work it, return full, and stream back through the corridors — and the contact rate rises for every ant they pass. When a patch is exhausted, the stream thins, and the rate falls.
The information about the colony's situation is real. It is in the rate — in the rhythm of contacts moving through the corridor, which is a faithful trace of what the whole colony has been doing. The ant reads that rhythm with its antennae and responds to it with its threshold, and in doing so it acts appropriately to a situation it has no picture of.
Remove the ant from the colony — put it in a dish, alone — and the same simple machinery that served it so well in the nest produces nothing. There are no contacts to count. There is no rate to read. The threshold sits there with nothing to measure against. The ant wanders, because the thing it was reading was never in the ant. It was in the colony the ant has been taken out of.
Nothing about the ant has changed. Its nervous system is intact. Its threshold is the same number it always was.
It would be easy to file this away as a fact about ants, a charming quirk of a small and distant animal.
Something just happened in the reader's own body in order to read that sentence.
A neuron in the reader's brain works the way the ant works. It sits, receiving signals arriving at its connections from thousands of other neurons. Each arriving signal nudges it a little. The nudges build up and leak away with time, as the ant's trace of recent contacts builds and leaks. And the neuron has a threshold. When the arriving signal climbs past it, the neuron fires, sending its own pulse onward. Below the threshold, it stays quiet.
Detect. Compare. Act. The same three operations. A neuron does not understand the thought it is helping to produce, any more than the ant understands the colony. There are roughly eighty-six billion of them in the reader's head, each doing only this.
And not only neurons. A cell in the reader's body deciding whether to divide is measuring the concentration of growth signals around it against an internal threshold. A bacterium deciding whether to swim toward food is comparing the concentration it senses now against the concentration it sensed a moment ago, and acting on the difference. In each, the same shape appears: sense something, compare it to a set point, do one thing or another depending on which side of the point you are on.
It is the same solution, arrived at again and again, because the problem is the same problem: how to make a single unit respond appropriately to a changing world using almost nothing.
It is the simplest computation that can be called a decision at all. A measured quantity, a threshold, and a switch. It is so simple that it seems impossible it could be the foundation of anything — of a foraging colony, of a beating heart, of a thought.
But that is the finding: this one computation, repeated, across thousands or billions of units at once, in a medium that carries the result of each unit's action to the next, so that what no single unit could compute is computed by all of them together.
The ant in the corridor is the thing itself, shown plainly, at a scale slow enough and large enough to watch. A rate, a threshold, a departure. Multiplied. That is what an ant is.
It is also what the reader is made of.