Collective ant behavior

How can an ant colony solve complex problems without a leader directing it?

There is a queen, so it is tempting to think the problem is already solved — someone is in charge. But watch what she does: she lays eggs. She issues no orders, inspects no routes, and could not see the foraging ground if she wanted to. Meanwhile the colony finds the nearest food, shifts effort when a better source appears, and abandons a trail when the food runs out. The decisions are real, and nobody is making them. How?

Shrink the problem. What can a single ant know? She is small, close to the ground, with a few centimetres of useful vision; she cannot survey the terrain or compare two routes she has not walked. The intelligence is plainly not inside her, so it must be in what passes between ants — and the medium is chemical. A returning forager drags her abdomen and lays a volatile trail; an ant crossing it tends to follow it, and if she too finds food she reinforces it on the way back.

Watch what that rule alone does. Suppose two routes lead to the same food, one short and one long, and ants set out on both at random. Which trail gets its second coat of pheromone first? The short one — simply because those ants get home sooner. A slightly stronger trail recruits slightly more followers, who strengthen it further, and before long nearly all traffic runs on the short route. No ant compared the two: the comparison was performed by the timing of return trips, and the answer accumulated in the chemistry of the ground.

That is the trick, but it has a flaw worth spotting. What stops the colony committing forever to the first decent trail it finds? Ask what pheromone does if nobody renews it: it evaporates. Forgetting is not a leak in the system, it is a component of it. When the food runs out the marks stop, the trail decays, ants revert to wandering, and the colony can find something new. Without evaporation it would be locked into its own history.

So there are two opposed forces — reinforcement, which sharpens a choice, and decay, which reopens it — and their balance sets how stubborn or flighty the colony is. The same shape reappears elsewhere with a different currency: ants also read encounter rate, so one repeatedly bumping into returning foragers effectively infers that foraging is going well and is likelier to go out herself. Nobody solves the problem; a population of simple responses amplifying and damping each other does.

Two corrections. First the queen: she is a reproductive organ, not an executive, and removing her ends the colony's future but not its foraging. There is no chain of command, and no ant holds a representation of the plan.

Second, "ants use pheromone trails" is broader than the evidence supports. Trail-laying is well documented in many species, but plenty of ants forage individually and navigate by sight, by the polarization pattern of the sky, or by counting their steps — desert ants of the genus Cataglyphis are the classic case and lay little or no trail. What survives across species is local rules plus amplification, not pheromone specifically. And the neat convergence on the shortest path is clearest in laboratory setups built to test it; in cluttered real terrain a colony often keeps several trails at once, which is less elegant and considerably more robust.

• How reinforcement-and-decay became a real optimization method in computing, under the name ant colony optimization.
• How honeybees choose a new nest site by a different mechanism, in which scouts actively inhibit one another.
• Why a colony's response to hunger is measured in encounter rates rather than orders.

Every term the collection defines is gathered in the glossary.