What Happened
Collective Behaviour · Biology · Source Verification · Source ReviewBBC Future · 11 Aug 2026
Executive summary · one-page brief
One Brake, Three Different Levers
A widely shared BBC Future feature draws five lessons in decision-making from ants, bees, locusts, cockroaches and fruit flies. Three of the five really do share something: cheap, parallel sampling, plus a brake that stops the group committing too early. But the brake is not one mechanism wearing three costumes — pheromone decay, a threshold-and-veto system, and a shrinking pool of active decision-makers are three different quantities doing a similar job. The other two cases are not about a brake at all: cockroach research is about how a mix of temperaments changes the speed of a group decision, and fly research is about one animal’s own hesitation, with no group in it anywhere. The freshest of the three real cases — a March 2026 paper the feature links but never names — is also the one that pins down exactly which quantity does the work.
By the Numbers three of five, one paper linked but unnamed
- 5insects the feature draws lessons from
- 3cases that are actually collective decisions
- 15outside sources the feature links — none of them named in its text
- 256people tested, in the experiment the feature calls “19”
- Mar 2026the paper behind the locust finding — linked from the feature, never named in it
What Is Established, and What Is Not graded by standing, not interleaved
| Standing | What | Source date |
|---|---|---|
| Confirmed | BBC Future’s feature names three researchers, in three of its five cases, and gives no paper, journal or publication year in its text for any of them — though it links its studies: 17 links to 15 distinct outside addresses | 11 Aug 2026 |
| Confirmed | The neutrality paper behind the locust finding is a real, peer-reviewed publication: Sontag, Hoffmann, Rogers & Yates, Advanced Science, DOI 10.1002/advs.202512301 | 22 Mar 2026 |
| Confirmed | The locust behaviour that paper analyses was filmed in 2006, by a separate team; the 2026 paper re-analyses that dataset rather than collecting new footage | 2006 & 2026 |
| Confirmed | The paper’s human experiment ran twelve replicate sessions with group sizes from 12 to 33 — 256 participants in total, not the single group of 19 the feature reports | 2026 (unpublished dates within the paper) |
| Confirmed | The cockroach study the feature links is a modelling study, as the feature says — simulations with no, some and a lot of personality variation, compared with experimental aggregation data; the group’s earlier experiments composed real insects, individually assessed as bold or shy, into groups | 2015, 2018 & 2020 |
Timeline
Forty Years Behind Five Paragraphs every dated node, in order
- 1992Marco Dorigo’s PhD thesis, at the Politecnico di Milano, turns the ant-colony metaphor into an optimisation method.
- 1998Di Caro & Dorigo publish AntNet in the Journal of Artificial Intelligence Research: forward and backward ant agents ride the same queues as real data packets.
- 2006Buhl and six colleagues film marching desert-locust nymphs in a ring-shaped arena, recording the density-driven switches in direction the 2026 paper will later re-analyse.
- 2010Thomas Seeley’s Honeybee Democracy (Princeton University Press) gives the bee-quorum account its name.
- 2011Burke & Waddell show fruit flies learn to prefer a nutritious sugar over a merely palatable one within minutes of eating it.
- 2011–12Seeley and five colleagues identify the cross-inhibition “stop signal” bee scouts use to silence a rival’s waggle dance, published in Science.
- 2014DasGupta, Ferreira & Miesenböck show a fly’s hesitation between two odours scales with how similar the odours are, matching a drift-diffusion model.
- 2015 & 2018Planas-Sitjà and colleagues show that cockroach groups mixing bold and shy individuals aggregate faster than groups of one temperament alone.
- Feb 2026The dataset behind the neutrality paper, and its voting-game code, is deposited at the University of Bath’s Research Data Archive.
- 22 Mar 2026Sontag, Hoffmann, Rogers & Yates publish “Consensus Formation and Change are Enhanced by Neutrality” in Advanced Science.
- 23 Mar 2026The University of Bath publishes its own press release on the paper, headlined “encourage opponents to sit on the fence.”
- 11 Aug 2026BBC Future publishes Sofia Quaglia’s “Want to get better at making decisions? Ask an ant”, naming three researchers and linking, without naming, the studies behind its five cases.
The Argument
The Same Brake — Or Is It? a shared function, three different mechanisms
Ants leave a pheromone trail that evaporates: without that decay term, a colony would lock onto the first path a scout happened to find, whether or not it was the shortest one. Bees run a quorum, but a quorum alone is not the brake — a second wave of scouts must independently corroborate a site before the swarm commits, and a scout that keeps dancing for a rival site sends a direct cross-inhibition “stop signal” that silences the other's dance, which is what stops one over-enthusiastic scout from carrying the whole swarm. Locusts do something different again: a swarm switches marching direction only after many individuals simply stop, and the 2026 paper's contribution is naming the quantity that stopping changes — the effective population size of individuals actually “voting” for a direction. Shrink that population and the same small, random nudge becomes a much larger share of it, which is what lets the group tip into a new consensus. Three quantities — a decay rate, a competitive-inhibition signal, and a population count — are doing a functionally similar job: each is a way of preventing a system from getting stuck on whatever it committed to first. They are not one mechanism run three times.
- decayants: a trail's strength falls over time
- vetobees: a rival scout's dance is actively silenced
- headcountlocusts: fewer active voters, more sway per vote
Three Cases, Three Mechanisms what actually varies, and what does not
| Case | Sampling | The brake |
|---|---|---|
| 螞蟻 Ants | scouts leave at random, laying pheromone as they go | Evaporation — a decay term with no counterpart in the other two cases |
| 蜜蜂 Bees | scouts dance in proportion to site quality; a second wave corroborates | Quorum + cross-inhibition — a rival's dance is actively vetoed, not merely outcompeted |
| 蝗蟲 Locusts | each individual aligns with its neighbours' direction of travel | Shrinking effective population — a headcount of active “voters,” not a signal between individuals |
- 01Samplecheap, parallel, random
- 02Reinforcesuccess feeds back
- 03Brakea different mechanism each time
- 04Commitonly once the brake releases
What Others Add
From Colony to Algorithm, and Back to a Contested Claim the lineage the feature does not give you
Ants · stigmergy
Trails That Forget, Turned Into Routing
- Marco Dorigo, director of IRIDIA (the AI lab at the Université Libre de Bruxelles), holds the PhD (1992, Politecnico di Milano) that first turned the ant-colony metaphor into an optimisation method.
- Di Caro & Dorigo's 1998 AntNet is the routing descendant: forward and backward ant agents ride the same queues as real data packets, so the routing table learns from the traffic actually there.
- Ant Colony Optimisation, the family AntNet belongs to, has, in the BBC feature’s words, been turned into an algorithm for improving human scheduling, telecommunications, transport and logistics, and has been used for railway planning — a lineage that runs from biology into engineering and back.
- No individual ant ever knows the route is optimal — the colony's decision emerges from evaporation acting on many ants' independent trips, not from any ant planning one.
Bees · quorum
The Name the Feature's Own Section Borrows
- The BBC feature's bee section names no researcher at all, unlike its sections on ants, locusts and cockroaches — the fly section is the only other one with no name in its text.
- The account it gives — scouts dance in proportion to site quality, a second wave corroborates, migration follows only once a quorum agrees — is the process Thomas Seeley’s Honeybee Democracy (Princeton University Press, 2010) is about: its publisher summarises the book as the story of how a swarm, choosing a new home, relies on collective fact-finding, vigorous debate and consensus building.
- Seeley and five colleagues later identified the mechanism that makes a quorum actually stick: a scout still dancing for a losing site receives a direct, physical “stop signal” headbutt from a scout returning from the winning one.
- That cross-inhibition is what stops a single enthusiastic scout from carrying the swarm — a quorum threshold alone, without a way to actively suppress a rival, would not do it.
Two Cases That Are Not About a Brake at All read against the paper, not the feature
Cockroaches choosing a shelter are a genuine collective decision — but the finding is not a brake, and runs, if anything, the other way. Planas-Sitjà and colleagues composed real cockroach groups of measured, consistent individual temperament (bold: quick to leave shelter; shy: slow to) and tracked which shelter a group settled on, over three consecutive days. Groups of only bold individuals moved around too much to settle at all; groups of only shy individuals settled fast, sometimes on a poor shelter; groups mixing both temperaments settled on a good shelter, and settled faster than either uniform group, because a mixed group generates more of the social interaction — amplification, correction — that a purely uniform group lacks. That is a finding about how the composition of a group changes the speed and quality of a decision it was always going to reach. It has no forgetting term, no cross-inhibition signal, and no shrinking population of active decision-makers in it anywhere.
Fruit flies are a different kind of case again: there is no group in the experiment at all. DasGupta, Ferreira & Miesenböck trained individual flies to discriminate between two odours and measured how long each fly took to choose. As the two odours were made more similar, each fly's own reaction time rose and its own accuracy fell, in a pattern that matched a drift-diffusion model — the same mathematical description used for a single human weighing evidence before answering a question. Flies carrying a mutation in the gene FoxP took longer to decide, at a given level of accuracy, than ordinary flies; a mushroom-body circuit that requires FoxP is where that decision is made. None of that is a swarm, a quorum, or a population of anything. It is one animal's own nervous system, deciding on its own.
- mix, not brakecockroaches: temperament mix changes speed, not commitment
- one animalflies: a single nervous system, no group at all
Conclusion
What to Take From Five Insects what transfers, what does not, what is still open
- Three cases share a function, not a mechanism. Ants, bees and locusts each pair cheap parallel sampling with something that prevents premature commitment — but the something is a different quantity each time: a decay rate, a cross-inhibition signal, a population count.
- Ground truth is missing from the human case. A colony converges on the shortest path because one objectively exists. Most human group decisions have preferences, not a provably correct answer — so the architecture (sample widely, delay commitment) transfers; a guarantee of finding the best answer does not.
- The neutrality finding is about changing a consensus, not forming one from scratch. Sontag, Hoffmann, Rogers & Yates test a swarm or a group switching an existing direction, not two groups arriving at a first decision. It is also symmetric: the same shrinking of active decision-makers that lets a deadlocked group finally settle is what lets a settled group be unsettled again.
- Two of the five cases are off-frame. Cockroach research is about how temperament mix changes the speed of a decision; fly research is one animal's own nervous system, with no group anywhere in the experiment.
- A widely read feature named three researchers and linked its papers without naming them. Following the links changes what the story is: not one mechanism explaining five insects, but three real mechanisms, doing a similar job — and a reader who does not click is never told which papers they are.
Bottom line
Good decisions under uncertainty come from cheap parallel sampling plus some mechanism that prevents early commitment. Pheromone decay, quorum-plus-veto, and a shrinking pool of active decision-makers are three different such mechanisms, not one restated three times — and the third of them was published in March 2026, in a paper the widely shared feature links but never names.
What to hold loosely
The neutrality mechanism cuts both ways: the same population-shrinking effect that lets a deadlocked group finally settle is, read the other way, a recipe for destabilising a consensus someone dislikes. Bath's own headline for the finding is exactly that operational reading — encourage opponents to sit on the fence.