Ant Colonies & Decentralized ConsensusPHEROMONE, EVAPORATION, QUORUM
How a crowd of nearly blind individuals arrives at a collective decision
Cited in: Topic 13 Self-Organization
01The Question It Poses
The queen gives no orders; her job is laying eggs. A single ant is nearly blind, holds no global picture, and has no idea how many tens of thousands of nestmates it has. And yet the colony finds the shortest route between nest and food, scales the number of workers on a task up and down with demand, and reaches agreement on where to move when it has to move.
Which makes the question sharp: where is the decision actually made? If not one of them knows the whole picture, what performs the thing that looks like a choice.
02The Rules
The foraging branch needs only three:
A forager wanders at random. Having found food, it walks back and lays pheromone (a volatile chemical trail) along the way.
On meeting a trail with pheromone on it, follow with higher probability. That probability rises with concentration but saturates — it is not linear.
Pheromone continuously evaporates.
The house-moving branch adds one independent rule, the quorum: once a scout meets more than a threshold number of nestmates at a candidate site, it switches from "keep evaluating" to "start carrying".
Notice what these rules do not contain: no comparison, no voting, no aggregation, nobody holding the two route lengths side by side.
Pheromone concentration on the two routes of a double bridge. The divergence is not decided by anyone; it is two positive feedbacks squeezing each other.
03What You See When It Runs
The classic double-bridge experiment (Deneubourg, Goss and colleagues, 1989–1990): two routes of unequal length between nest and food.
At first the ants split at random. The ones on the short route complete a round trip faster, so they lay more pheromone per unit time, so concentration on the short route rises faster, so later ants are more likely to take it — feedback that widens the gap. Within tens of minutes almost everyone is on the short route. No ant compared the two. The comparison was done for them by the physical fact that a faster round trip deposits faster.
Change the experiment and you see the cost. Open only the long route, let the colony lock onto it, then open the short one. The colony often does not switch — concentration on the long route is already too high for a thin new trail to compete. That is not a defect in the model; it is the built-in price of this mechanism: positive feedback buys convergence and sells the ability to turn back.
What recovers part of it is rule 3. The evaporation rate is the knob: fast evaporation makes the colony responsive to change but prone to dithering between routes and losing efficiency; slow evaporation gives clean convergence but no capacity to change its mind. Neither end is "better" — they are two ends of one trade-off.
The same environmental change (a shorter route opening midway) gives completely different outcomes for fast and slow evaporation.
04What It Explains
In engineering the mechanism was lifted wholesale: Ant Colony Optimization (ACO), from Marco Dorigo's 1992 doctoral thesis, uses artificial pheromone plus artificial evaporation to attack routing, scheduling and travelling-salesman-type combinatorial problems. Its strength is not that it finds the best answer, but that it doesn't need restarting when the problem changes while it runs.
In collective decision-making, ants of the genus Temnothorax use a quorum to choose a new nest. The interesting part is that the threshold is adjustable: experiments by Franks and colleagues in 2003 showed that when the old nest has been destroyed and the colony must move immediately, it lowers the quorum — trading accuracy for speed. That turns "the speed–accuracy trade-off" from a slogan into a parameter you can manipulate experimentally.
In division of labour, the response-threshold model gives an equally plain account: each ant has its own threshold for a class of stimulus (hungry larvae, a breach in the nest wall) and gets to work once the stimulus exceeds it. As more ants work, the stimulus falls, and the high-threshold ones stay put. No scheduler, and worker numbers still track demand.
What It Cannot Explain
It guarantees nothing optimal. Locking onto a suboptimal route is normal, especially after the environment has changed. ACO controls this in practice with extra machinery (elitist strategies, upper and lower bounds on concentration) — machinery with no counterpart in real ants. Engineers added it.
It doesn't represent all ants. A great many species lay no pheromone at all: the Saharan Cataglyphis navigates by path integration (keeping a running vector sum of where it has walked) and visual landmarks, and can run straight home from hundreds of metres out. "Ant colony = pheromone" takes a handful of species as the whole taxon.
It doesn't explain colony-level change. Decades of field work by Deborah Gordon on red harvester ants show colony behaviour shifting with colony age — older colonies are steadier and overreact less to disturbance. Individual workers live about a year. That memory is in no individual ant, and it is not in the pheromone model.
Working well in engineering is not biological validation. ACO solving problems well says nothing about how ants behave. All the two share is the skeleton of positive feedback plus forgetting; most of the details that make the algorithm actually good have no biological counterpart.
It does not transfer to people. Pheromone is a costless, intentionless trace: an ant cannot forge it, cannot hide it, and cannot lay extra to send a nestmate the wrong way. Every trace people leave — ratings, likes, citations, résumés, transaction prices — can be manufactured strategically. So the first question when applying this mechanism to a human system is: can the trace be forged? If it can, what feedback amplifies is not quality but the ability to forge.