Five Insects, One Mechanism

A BBC feature offers five lessons in decision-making from ants, bees, locusts, cockroaches and fruit flies. The lessons are not five — they are one, wearing five costumes. Ants forget their trails, bees wait for a quorum, locusts pause, cockroaches differ from one another, flies hesitate longer when the choice is close. Every one of those is a brake on premature convergence: cheap parallel sampling, plus something that stops the group committing too early. That single idea is more useful than the five it is split into — and the freshest result behind it, a March 2026 paper the article never names, identifies exactly which quantity does the work.

By the Numbers what the piece contains, and omits

5insects the article draws lessons from
3that are actually collective decisions
1998the AntNet paper the translation turned into a firm
Mar 2026the locust result — never cited in the piece

The Five, and What Each Really Is three are collective · two are not

Case	Feedback	The brake on early commitment
螞蟻 Ants	pheromone reinforcement	Evaporation — a forgetting term. Without it the colony locks onto the first path it finds.
蜜蜂 Bees	dance recruitment, scaled by site quality	Quorum + verification — a second wave checks, so one enthusiastic scout cannot carry the swarm.
蝗蟲 Locusts	alignment with neighbours	The pause. Here the variation term is not noise to be minimised — it is the control lever.
蟑螂 Cockroaches	— not consensus dynamics	Heterogeneity as a hedge. All-bold groups never settle; all-timid groups settle fast on a poor site.
果蠅 Fruit flies	— individual cognition	Difficulty-scaled hesitation. Closer odours, longer deliberation. Not a swarm study at all.

The freshest result is the one the article leaves anonymous, and it is worth stating properly because it inverts the intuition. Neutral participants are not merely harmless: they shrink the actively-voting population. In a large, fully committed group each individual's small independent influence is cancelled by everyone else's, and that cancellation is exactly what makes a settled position immovable. Let people stand aside and the effective pool shrinks, so the same absolute nudge becomes a larger share of it — the system turns noisy, and noise is what lets it move. In the locusts the evidence is unusually clean: every switch of marching direction is preceded by a spike in the number of individuals that have simply stopped.

fewer votersthe same nudge, a larger share of a smaller pool
pause firstthe swarm turns by stopping, not by arguing

Bottom line

Good decisions under uncertainty come from cheap parallel sampling plus a mechanism that prevents early commitment. Evaporation, quorum, pause, diversity, hesitation — five names for one brake.

The strain

Ants and bees solve problems with a right answer. Most human group decisions have none — so the architecture transfers, the guarantee does not.

起承 · Context · Content 1 / 2

起 — How a Colony Decides Without a Decider the two classical cases

Ants · stigmergy
Trails That Forget
shortest path, no plan
- Scouts leave randomly, with no route constraint at all.
- Each lays pheromone going out and coming back; ants prefer the strongest scent.
- A shorter path is walked more often per unit time, so it accrues more scent.
- Pheromone evaporates — the forgetting term that prevents lock-in.
- No individual ant ever knows the route is optimal.
Bees · quorum
A Dance With a Threshold
advertise, verify, commit
- Scouts return and dance rather than leaving a trace.
- Better site → longer and more vigorous dance. Quality sets amplitude.
- The dance encodes distance and direction, so it is checkable.
- A second wave flies out to verify — a built-in replication step.
- Migration only once a quorum agrees. The threshold is the commitment device.

承 — From Colony to Algorithm where it left biology

Ant Colony Optimisation is the direct descendant, and it is in production: staff rostering, telecommunications, transport and logistics, and railway scheduling. Marco Dorigo, co-director of the AI lab at the Université Libre de Bruxelles, puts the transferable claim precisely — efficient collective decisions can emerge without central control, from simple individuals with limited capability and limited information. The routing application, AntNet, is a 1998 paper by Gianni Di Caro and Dorigo in the Journal of Artificial Intelligence Research: forward and backward ant agents ride the same queues as real data packets, so the routing table learns from the traffic actually there rather than from a model of it.

1998AntNet — Di Caro & Dorigo, JAIR 9:317–365
not a firmthe BBC translation turned that surname into a company

The Lineage the Article Does Not Give You every one of these is citable

1992Dorigo's PhD thesis at Politecnico di Milano turns the ant-colony metaphor into an optimisation method.
1998Di Caro & Dorigo publish AntNet in JAIR — forward and backward ant agents ride the same queues as real data packets.
2010Seeley's Honeybee Democracy (Princeton) supplies the account, and the word, that the article's bee section borrows unattributed.
Mar 2026Sontag, Hoffmann, Rogers & Yates name the quantity doing the work: effective population size.
Aug 2026The BBC feature: five lessons, three researchers named, zero citations.

Three Gaps in the Source checked, not assumed

Gap	What the article printed	What is actually the case
Invented company	「電子製造服務多利歐（Dorigo）公司則開發了蟻網」	AntNet is a paper by the professor quoted four paragraphs earlier — no such company exists
Unnamed paper	「最近一項關於蝗蟲的研究」, Yates credited alone	Sontag, Hoffmann, Rogers & Yates, Advanced Science, March 2026 — four authors, no citation given
Missing source	「民主蜜蜂」 framing, no researcher named	Thomas Seeley's work, and the 2010 book that put the word democracy on it

轉合 · Pivot & Resolution · Content 2 / 2

轉 — Noise Is the Resource the result the article did not land

The finding is not that neutral people are nice to have. It is that neutral actors reduce the effective population size during a transition. In a large, fully committed group every individual's small independent influence is cancelled by everyone else's; the aggregate is stiff, and that stiffness is precisely what makes a settled position hard to shift. When many step out and become neutral, the actively voting pool shrinks — so the same absolute nudge becomes a larger fraction of it, the system turns noisy, and a noisy system can be knocked into a new basin. The locust evidence is the clean part: in an annular arena, with every nymph classified at every moment as clockwise, anticlockwise or stopped, each direction switch is preceded by a spike in the stopped fraction. The swarm does not turn by arguing. It turns by pausing. The mechanism is also, unavoidably, symmetric: the same reduction in effective population size that lets a deadlocked group finally settle is what lets a settled group be unsettled. Bath's own announcement puts the operational version of that bluntly — encourage your opponents to sit on the fence. Read as advice for a team that cannot decide, it is a way of manufacturing the breathing room to reassess. Read as advice for a motivated actor facing a consensus they dislike, it is a lever for destabilising one. The article prints only the first reading.

fewer votersthe same nudge, a larger share of the pool
pause firstevery locust direction switch begins with stopping

合 — The One Brake, in Five Costumes sample widely · commit late

01Samplecheap, parallel, random
02Reinforcesuccess feeds back
03Forgetevaporation
04Pausego neutral
05Diversifybold and timid
06Commitonly at quorum

Where the Analogy Is Weaker Than Presented stated, not buried

Ground truth is missing. A colony converges on the shortest path because one exists. Most human group decisions have preferences, not optima.
The human sample is small. The article reports 19 participants in the voting game; the press material does not state a number, so it is uncorroborated here.
The finding is symmetric, and that cuts both ways. Neutrality makes consensus easier to form and easier to break.
Bath's own headline is a manipulation recipe — "encourage opponents to sit on the fence". The article gives only the constructive reading.
Two of the five are off-frame. Cockroaches are about heterogeneity; fruit flies are individual cognition, not a swarm at all.
Follow the citation — which is the one thing this article does not let its readers do.
The optimum can be the wrong target. Pheromone evaporation exists because converging on the best path found so far is a trap, not a goal.
Borrow the architecture, not the guarantee. Parallel sampling and a late commitment threshold transfer; "the group reliably finds the optimum" does not.
Check a claim where it enters the record. The invented company would have cost one search to catch, and it survived the review the page advertises.

Where it lands

The science underneath is real, and in the locust case genuinely new. The piece fails at exactly one job: it never tells you what to read next — no paper titles, no co-authors, no year, and Seeley absent from a section that is entirely his.

The irony, precisely

An article arguing that we over-trust our own deliberation carries a machine-translated error that invented a company from a surname — and it passed the human review the page advertises. The failure was not a lack of thinking. It was the absence of a cheap check where a claim entered the record. Ants solve that with evaporation.