And when it is very much stupider. A crowd can weigh an ox to within a percent, route a delivery fleet, and choose a nest better than any ant in it — or it can stampede, bubble, and lock itself into a certainty that no single member ever held. The difference is not intelligence. It is whether the members are still thinking for themselves.
Galton went to a livestock fair in 1906 expecting to prove that crowds were stupid. Eight hundred people guessed the weight of an ox. He took the median. It was off by less than one percent — closer than any cattle expert in the room.
The mechanism is not magic; it is arithmetic. Each guess is the truth plus an error. If those errors are independent, they point in every direction and cancel. What survives is the signal. The error of the average shrinks like 1/√N — to halve it you need four times the crowd.
Every word of that depends on one fragile assumption: that nobody looked at anybody else's answer.
Now let them answer in turn, in public. Each person still gets a private hint — right, say, 65% of the time. But they can also see what everyone before them chose.
Two people happen to guess A first. The third has a private hint saying B, but two votes say A — so she rationally discards her own information and says A. Now it's three to nothing. Everybody after her does the same. Nobody lied, nobody was foolish, and the crowd's private knowledge is never spoken aloud. This is an information cascade, and it locks in whichever way the first two coins fell.
Compare the two bars: what the crowd knows, and what the crowd says.
A colony of Temnothorax ants must choose a new nest. Scouts find candidates and recruit — but a scout only commits when she has personally counted enough other ants already inside. That count is a quorum.
The quorum is a dial between two failures. Set it low and the colony decides fast, on the first site anyone liked — a cascade with legs. Set it high and every site is thoroughly assessed by many independent scouts, and the colony chooses well, slowly. Ants raise the quorum when they have time and drop it when the roof is coming off.
Speed and accuracy are not both available. The colony does not solve this. It tunes it.
Averaging is not the only thing a crowd can do. Set ants loose on a set of cities: each walks a tour, biased toward short hops and toward paths that smell of previous ants. Afterwards, the ones who found short tours lay down more scent. Scent evaporates.
Nothing here holds a map. No ant compares two tours. Yet the pheromone field slowly becomes a memory of what worked, and the colony converges on a route far shorter than any single ant could plan. This is ant colony optimisation, and it is used to route real delivery vans.
Watch the faint paths thicken. That is the colony thinking.
Two hundred people, one binary question, and a truth only the simulation knows. Each person gets a private hint — right more often than not. A ring shows what they were told; the fill shows what they said. When the two disagree, someone knew better and said nothing.
Answer alone and the room is almost never wrong — and almost never agrees. Answer in turn, in public, and watch the cascade: after four or five speakers the tally leads by two, everyone rationally follows it, and the room goes unanimous — often unanimously wrong. Then set a quorum, set a deadline, and find the answer the ants found.