Ten Thousand Bees, No Boss

how a crowd with no leader, no ballot, and no bee that has seen two options at once still picks the right house

Jump to the simulation: a swarm choosing a home

Some warm afternoon in late spring, a third of a colony’s bees pour out of a hive all at once, swirl into a dark cloud, and settle onto a nearby branch as a hanging clump the size and shape of a football — thousands of bodies gripping each other’s legs, the old queen buried somewhere in the middle. They have walked out on their home and taken almost nothing with them. They have a few days’ honey in their stomachs and a problem that has to be solved before the honey runs out or the weather turns: somewhere within a few miles there is a hollow tree, or a gap in a wall, or an empty box, that will be the difference between a colony that survives the winter and one that freezes or starves. They have to find it, agree on it, and fly there together, as one body, to a place not one of them has ever seen. And there is nobody in charge.

Sit with how strange that is. There is no foreman bee, no committee chair, no architect holding the blueprint. The queen is not deciding; she is being carried. The cloud of bees on the branch contains no individual who knows the neighborhood, weighs the options, and issues the order. Worse than that: there is no bee anywhere in the swarm who has even seen two of the candidate homes to compare them. The information needed to make the choice is scattered across hundreds of separate little heads, each holding at most one piece of it, and somehow those pieces have to be combined into a single good answer that the whole crowd will bet its life on. If you handed that problem to a human organization, you would build a hierarchy to solve it — surveys, a manager, a meeting, a vote. The bees have none of that. And they are better at it than the committee.

Here is what you would expect to happen, if you know how crowds usually behave. Whichever house gets championed first should win. The earliest scout to come home dancing draws a few recruits; those recruits draw more; the crowd snowballs, and the swarm locks onto whatever it happened to see first — not because it is good, but because it had a head start. Rich gets richer. It is the same trap that crowns whichever stock is already hot, whichever video is already trending, whichever idea got loud early. A leaderless vote ought to be the easiest thing in the world to stampede.

And it nearly is. But Thomas Seeley spent the better part of forty years — on the ground, and on a windswept island in Maine where he could control every cavity for miles — showing that the swarm does something stranger. Choosing among a dozen possible homes that differ in volume, entrance size, height, and dryness, it reliably settles on the best one of the lot. Not the one found first. Not a decent one. The best one. It can abandon a site that got an early crowd going and swing the whole consensus to a better house discovered late. There is a single trick that lets it do that — one mechanism that defeats the head-start — and it is, as it happens, the same trick running inside your own skull when you make up your mind.

Two signals and a head-count

The work is done by a few hundred of the older bees, the scouts. While the rest of the swarm hangs quietly on the branch conserving energy, the scouts fly out in all directions and go house-hunting. A scout who finds a promising cavity crawls all through it, pacing the walls, measuring the volume by walking, checking the entrance and the height and whether it is dry. Then she flies home to the clump and does something remarkable: she argues for it.

She argues in the one language bees have for talking about places — the waggle dance, the same figure-eight shimmy a forager uses to point hivemates at flowers. The direction she waggles encodes the bearing to her cavity; the length encodes the distance. But here is the hinge of the whole thing: she dances harder for a better site. A scout returning from a truly excellent cavity will throw herself into the dance with real vigor, repeating the circuit again and again; a scout back from a so-so hole does a few half-hearted laps and quits. The quality of the real estate, which lives out there in the world, gets translated into the number of dance circuits, which lives here on the branch where other bees can feel it. Better house, longer dance. That is signal one.

And a dance is an advertisement. Uncommitted scouts — bees who haven’t found anything yet, or who’ve come home from a dud — bump into the dancers, read off the directions, and fly out to inspect what the fuss is about. If they like what they find, they come back and dance for it too. So support for a cavity feeds on itself: a few dancers recruit more visitors, who become more dancers, who recruit still more. This is positive feedback, the same snowballing engine that drives a crowd into a chant, a rumor through a town, or a stock into a bubble — and on its own it is exactly the rich-get-richer trap from a moment ago. Whichever site builds a crowd first builds it fastest, and the swarm slides toward the early favorite whether or not it is the good one. Recruitment is the swarm’s accelerator. By itself, it has no steering.

Worse, positive feedback alone has a famous failure of its own. If two cavities are both pretty good, each grows its own snowballing crowd of dancers, and the swarm locks into a stalemate — two big factions, neither backing down, the decision frozen. For an animal that has to lift off as a single body, a tie is not a harmless draw; a swarm that splits its vote can split in the air and lose its queen. So an accelerator is not enough. The swarm needs something that can overrule the head start — pull support back off a site that merely got loud early, and break a tie before it strands the colony. That something is the second signal, and it is the heart of the whole story.

A scout committed to one cavity will seek out scouts dancing for a rival cavity and deliver a sharp little head-butt paired with a brief vibrating beep — a signal that lasts about a sixth of a second. Seeley’s group called it the stop signal, because that is what it does: a dancer who takes enough of these head-butts falls silent and drops out, reverting to an undecided bee. Crucially, a scout aims her stop signals at the supporters of other sites, never her own. So the more support a cavity gathers, the more stop signals its crowd rains down on the competition — and a rival, being smaller, can’t answer in kind. This is cross-inhibition: each option actively suppresses the others in proportion to its own backing. And notice what it is the opposite of. Recruitment is rich-get-richer; cross-inhibition is the brake that undoes rich-get-richer. It does not reward whoever got loud first — it lets the swarm tear support away from an early front-runner and hand it to a latecomer, so the site that simply danced soonest does not get to keep the lead it never earned. The stalemate breaks not because one faction tires, but because the better-backed site actively silences the other. The accelerator finally has a steering wheel.

One rule decides when to stop arguing and go. A scout standing at a cavity keeps a rough count of how many other scouts are bustling around the same entrance, and when that local crowd passes a threshold — a quorum of fifteen or twenty bees at one hole — she reads it as “we’ve agreed,” flies home, and helps warm up the swarm for takeoff. That is the entire machine: dances that grow with quality to gather evidence for each option, stop signals that let the leading option suppress its rivals, and a quorum that calls the question once one option is sure enough. No bee ever tallied the whole vote. No bee compared the options. Each one knew only her own cavity and the size of the crowd standing in it, and out of that the swarm computes a choice. Below, you can run the argument yourself — and you can switch the stop signal off and watch the rich-get-richer trap close.

The Experiment

Things to try:

What you have been watching has a name in other fields, and the name is worth saying plainly, because it is the punchline. The two signals — dances that build support for an option in proportion to its merit, and stop signals that let each option suppress its rivals — are a machine for integrating evidence. Each cavity is a hypothesis; each dancing scout is a scrap of evidence in its favor; the crowd at a site is the running total of evidence accumulated so far; and the quorum is the moment the total crosses a confidence threshold and the swarm acts. The bees are not voting. They are computing, the way a good instrument computes — piling up noisy evidence for competing answers until one answer is sure enough to bet on.

The swarm is not a crowd that happens to choose. It is a brain that happens to be made of bees.

The same trick, in your head

That last line is not a flourish. When Seeley laid his swarm data beside what neuroscientists had measured inside a primate brain making a simple choice — which way is that patch of dots drifting, left or right? — the two pictures turned out to be the same picture. In the brain, rival populations of neurons each pile up evidence for one answer, each firing faster as the evidence for its option mounts; each population inhibits the others; and when one population’s firing crosses a threshold, the decision is made and you act. Swap “neuron” for “scout” and “firing rate” for “dancing crowd” and you have the honeybee swarm, exactly. Cross-inhibition between groups of evidence-gatherers, racing to a threshold, is apparently just how you make a good decision out of unreliable parts — whether the parts are cells in a skull or bees on a branch.

This is the deep thing this whole site keeps circling. A capability we are tempted to credit to some central self — deciding, that most first-person of acts — turns out to need no center at all. It is a pattern that runs on any pile of simple units that can vote up and shout down. Your neurons do not know they are in a brain; the scouts do not know they are in a swarm; and in neither case does the intelligence live in a part. It lives in the wiring between the parts. You have met this idea elsewhere on the site — in the way a flock turns as one with no lead bird, in the way a slime mold finds the shortest path with no map — but here it lands closest to home, because the thing the swarm is doing is the thing you most think of as yours.

The thing the bees got right

It would be easy to end on a warm note about the wisdom of crowds, and easy is exactly the trap. The honest reading of this page is sharper, and it cuts the other way too. A crowd locked in positive feedback, each member copying the excitement of the others, is not automatically wise. It is the identical machinery that runs a panic through a market, a stampede through a crowd, a bubble through a herd of investors all dancing for the same hot stock because everyone else is. Recruitment that feeds on itself will just as happily carry a swarm onto a terrible site, if the dancing comes loose from the quality of the thing being danced about. A mob is a swarm with the wiring crossed.

So the question worth carrying off is not “why are crowds smart” — they often aren’t — but what the bees got right that mobs get wrong. Two things, and they are both about honesty. First, a scout’s enthusiasm is yoked to something real: she dances hard only for a cavity she has personally crawled through and measured, so the loudest signal in the swarm is, on average, tracking the truth and not just the volume. Second, the stop signal gives every option a way to puncture the runaway excitement of its rivals, so no single contagion gets to grow unopposed. Honest advertising, and a built-in brake on hype. Strip either one out — let the dancing detach from real quality, or remove the cross-inhibition that lets good sense interrupt a craze — and the same swarm that reliably finds the best home becomes a herd pouring confidently off a cliff. The universe hands out the crowd-machine to bees and to mobs alike. It does not come with the safeguards attached. Those, the bees had to evolve, and those, we have to build.

There is one last thing to notice, and it is the quiet center of the whole spectacle. At no point in this process does any bee hold the goal. Not the queen, dangling passively in the warm interior. Not the scouts, each of whom knows a single hole in a single tree and a rough sense of how many of her sisters are standing in it. Nowhere in the swarm is there a representation of “find the best home and move us into it.” That purpose exists only in the swarm as a whole, the way a melody exists only in a run of notes and in none of them. And yet the purpose is unmistakably there — in the reliability, the accuracy, the way the thing gets the answer right far more often than chance. A want with no wanter; a decision with no decider; an intelligence with nothing inside it that is intelligent. It hangs from a branch in somebody’s orchard every spring, and then one bright morning it lifts off all at once and flies, with perfect agreement, to a place none of them has ever seen.