What Forty Years of Engineering Never Taught Me About Deciding Things

When a honeybee colony outgrows its hive, roughly half the bees leave with the old queen in a swarm and cluster on a branch somewhere nearby, a trembling ball of ten or fifteen thousand insects with nowhere to live, and what happens next is, I think, the most astonishing decision-making process I have ever studied, and I say this as a man who spent thirty-five years designing control systems for hydroelectric turbines, where every decision had an owner and a signature and a chain of responsibility running all the way up to someone in an office who could be blamed if it went wrong. Bees have no such office. Nobody is in charge. And yet they get it right, most of the time, better than most of the committees I sat on.

Here is what actually happens, and I want to walk through it carefully, because the mechanism matters more than the metaphor.

While the swarm clusters, a few hundred of the older, more experienced bees become scouts. They fly out in all directions, sometimes for kilometres, searching for a suitable new home — a hollow tree, a chimney, a wall cavity, something dry, defensible, with an entrance small enough to guard and a volume of roughly forty litres, which the bees seem to know without anyone having told them. When a scout finds a candidate site, she inspects it thoroughly, walking its interior surfaces, and then flies back to the swarm and performs what is called a waggle dance: a figure-eight movement in which the angle of the straight run encodes the direction of the site relative to the sun, and the duration of the waggling encodes the distance. This part was worked out by Karl von Frisch decades ago and is by now textbook biology. What is less well known, and what I find more remarkable, is what the dance also encodes: enthusiasm. A scout who has found a mediocre site dances briefly, a handful of circuits, and stops. A scout who has found an excellent site — dry, well-ventilated, the right size, a small defensible entrance — dances vigorously, for many minutes, sometimes returning to the site and coming back to dance again and again, recruiting other scouts to go check it for themselves.

This is the part that undid my assumptions about how good decisions get made. I spent my career believing that good decisions require a single evaluator with complete information — an engineer with the full drawing, a manager with the full report. The swarm has no such evaluator. What it has instead is many independent scouts, each visiting only one or two sites, each reporting honestly through the vigour of her dance, and — this is the crucial part — other bees do not simply trust a dance and go home. They watch it, then fly out themselves to verify the site independently, and only then dance for it themselves if they agree. A weak site gets weak dances that fade out within a day, because no second bee bothers to confirm it and re-advertise it. A strong site accumulates more and more dancing scouts, each of whom checked it herself, and the growth is not linear but compounding — success recruits success — until a threshold is crossed. That threshold has a name, quorum sensing, and researchers have measured it directly: when roughly fifteen to twenty scouts are present at the same site simultaneously, the decision tips. Those bees return to the swarm and begin a new signal, called piping, a short vibrating buzz against the other bees' bodies, which is the instruction to warm up their flight muscles. Within an hour or two of quorum being reached, the entire swarm — ten thousand bees who never visited the site themselves, who are relying entirely on the aggregated judgement of a few hundred scouts they will never individually verify — lifts off together and flies, in a loose cloud with the fastest scouts weaving through it to steer the slower bees, directly to the chosen site. They are right, by the measurements researchers have done comparing chosen sites against the alternatives that were scouted but not chosen, on the order of ninety percent of the time.

I have sat in perhaps four hundred engineering meetings in my career, and I would guess, conservatively, that in a third of them the room reached consensus not because the best option had been independently verified by multiple people but because the loudest or most senior voice in the room had decided early and the discussion afterward was a search for reasons to agree. The swarm cannot do this. There is no senior bee. A scout with a mediocre site cannot out-dance a scout with a good site through force of personality; her dance simply fades because nobody re-confirms her site, while the good site accumulates independent verification precisely because it is good. The system is resistant to persuasion and vulnerable only to genuinely misleading evidence — a site that looks good on first inspection but has a flaw the scout missed. This is, as far as I can tell, functionally identical to what statisticians call a wisdom-of-crowds mechanism with independent sampling and weighted aggregation, except it evolved several million years before anyone wrote that phrase down, running on insects with a brain the size of a grass seed.

I do not want to overstate the parallel to human institutions; bees are not people, and the stakes of a bad hive site are not the stakes of a bad reservoir design. But I have kept bees for five years now, since my retirement, and every autumn when I watch a swarm cluster and disperse and choose, I find myself doing something I never did in thirty-five years of engineering meetings, which is trusting a process I cannot fully see, because I have watched it, season after season, arrive at good answers without anyone in charge of arriving at them. I do not know what to do with that, professionally, at sixty-five. I only know that I have started, in my own small decisions, asking a question I never asked as an engineer: not who is right, but how many independent people have actually gone and checked.

Read in the app, leave a resonance, discover more stories.