A hive and a cortex, described the same way

Generated from this site's git history. with Claude means the commit carries a co-author line for an Anthropic model: an LLM took part in that change, not necessarily in every word of it, and git cannot say which. The same history on GitHub →

1. v2 2026-09-12 07:54 posts: inline source cards, and every post cut to the point Joshua Anderton with Claude Opus 5 +580 −1087 words

   description: >-

   Two systems — aA honeybee colony choosing a nest and a cortical population choosing between

   choosing between two motions — have been givencarried the same formal description since 2009. What it

   since at least 2009. What that description covers, and where its own authors say it stops.

   date: 2026-09-12 07:00:00 -0600

   tier: C

   inline_sources: true

   claims:

   - barron2007

   - linn2020

   - dong2023

   - linn2020

   - cardosojunior2020

   - myersjoseph2024

   citation_notes:

   - marshall2009b >> The formal treatment. Read at full text for the sections carrying its conclusions; the model derivations were not read.

   - marshall2009 >> Same authors, same year, the idea under a different heading.

   - seeley2012 >> The experimental result usually paired with the framework above.

   - reina2017 >> Extends the analysis past two options.

   - lin2026 >> Results section read in full. The statistics quoted here come from it; no interpretation is attributed to it.

   - myersjoseph2024 >> Published version of a preprint; the published title narrows the claim to cross-modal attention.

   image: /assets/posts/waggle-dance.jpg

   summary: >-

   A swarm deciding where to nest and a cortical population deciding which way

   dots are moving have been modelled with the same equations for over fifteen

   years, and the idea is older than that. This is an inventory of what the

   correspondence actually covers, which parts of it are the authors' own claim

   rather than an inference, and the two places the papers themselves say it

   stops.

   Inventory of an existing parallel, not a proposal. A swarm choosing a nest and

   a cortical population choosing a direction are modelled as the same

   statistical procedure. The parallel is not mine; it dates to at least 1979 and

   was formalised in 2009.

   ---

   This is an inventory, not an argument. The parallel described below is not mine and is not new; the point of writing it down is to be precise about how much of it is established, and where the edges are.

   An inventory. The parallel below is not mine and is not new.

   ## The observationparallel, and whose it is

   A honeybee swarm choosing between two candidate nest sites and a cortical population choosing which way a field of dots is moving have been given the same formal description. The clearest statement of it is a 2009 paper in *Journal of the Royal Society Interface* {% include cite.html key="marshall2009b" %}, which sets a model of primate cortical decision-making beside three models of house-hunting in ants and honeybees and works through what they share.

   Its own opening is careful to say the idea was already old: speculation about whether decision-making in brains and in insect colonies is related, in the authors' words, begins "at least with Hofstadter (1979)" and had been taken up repeatedly before they wrote. They describe their contribution as a first step toward a common theoretical framework — one based on the interactions between a system's components rather than on what the components are. A companion piece the same year discusses the same territory under the heading of colony-level cognition {% include cite.html key="marshall2009" %}.

   A swarm choosing between two nest sites and a cortical population choosing which way dots move have been given one formal description.

   {% include source.html key="marshall2009b" %}

   That paper dates the idea to "at least Hofstadter (1979)" in its own introduction, and describes itself as a first step toward a common framework — one built on how a system's components interact rather than on what they are.

   {% include source.html key="marshall2009" %}

   So the lineage runs roughly: an analogy in the 1970s, recurring discussion, a formal treatment in 2009, and then theThe experimental work mostusually oftenpaired cited alongsidewith it, onconcerns inhibitory signalling between scouts. {% include cite.html key="seeley2012" %} — a paper whose ownIts title and framing put the comparison to brains in front and centre, rather than leaving it to a reader.

   {% include source.html key="seeley2012" %}

   ## What the correspondence actuallydescription covers

   Three thingsproperties, as set out inper the 2009 treatment:

   - **Populations, not individuals.** Both systems are modelled as mutually interactingInteracting populations that accumulate noisy evidence for competing alternatives.

   - **A threshold on one population.** AThe choicealternative is made when onewhose population's activity crosses a threshold, and the alternative that population representsfirst is the one taken.

   - **A threshold that can be moved.** The same threshold trades speed against accuracy. Raise it and decisions get slower and more reliable; lower it and the reverse. In the cortical model this is a parameter; in the ants it is the number of individuals that must accumulate at a site.

   - **A threshold that moves.** Raising it buys accuracy and costs speed. A parameter in the cortical model; a headcount at a site in the ants.

   The formal claim attached to this is stronger than a resemblance. Under particular settings, the cortical model reduces to a diffusion process that implementsimplementing the sequential probability ratio test — the procedure that, for a two-way choice, minimisesthe procedure minimising decision time forat any given error rate. The 2009 analysis asks which of the colony models can do the same thing, and treats the answer as a source of testable predictionsprediction about how colonies ought toshould behave if they are deciding optimally. That is the substance of the parallel: not that a hive resembles a brain, but that both can be written as the same statistical procedure.

   Extending the same analysis past two options changes it. The best-of-N case — more than two options — has been analysed separately for whether the controlling parameter stays the same {% include cite.html key="reina2017" %} — which is also the point where this literature meets the best-of-N problem in distributed and robotic systems, since that is the same problem stated without the bees.

   Past two options the analysis changes.

   {% include source.html key="reina2017" %}

   This is also where the literature meets the best-of-N problem in distributed and robotic systems — the same problem stated without the bees.

   ## Where theits papersown authors say it stops

   Two limits come from the 2009 treatment itself, not from a critic.

   **The colony has tomust go and get its informationevidence.** In the cortical task, evidence for both alternatives arrives at a fixed, equal rate. A colony's scoutsScouts discover sites stochastically, which leaves the colony holdingleaving an explore-or-exploit problem the cortical version does not have: assess thewhat sites alreadyis known, or look for better ones. The authors describecall this asit a hybrid of a bandit problem and a minimum-decision-time problem, and say no onenobody had formalised it.

   **Consensus andis not foraging are different problems.** This one is easy to lose. The paper states plainly that the decision a colony solves while emigratingemigration is *optimal consensus* decision-making, and that this "differs from the problem of distributed resource intake maximization that colonies tackle during foraging." The quorumQuorum figures, and the cross-inhibition result, come from nest-site selection during swarming. Foraging recruitment isruns governed byon different machinery. A description that moves freely between the two is describing two systems, not one.

   ## Two directions the mapping is being pushedextensions, neither established

   The following are working framings, recorded here as untested rather than as findings.

   **A modulatory layer.** The decision models above have recruitment and inhibition in them, and nothing that sets the regime those operate in. Bees have candidates for such a layer: octopamine dosing has been examined for its effect on how strongly value is reported in dances {% include cite.html key="barron2007" %}, octopamine and dopamine have been examined in dance following and information use {% include cite.html key="linn2020" %}, and an inhibitory signal associated with danger has been examined against dopamine levels {% include cite.html key="dong2023" %}. Whether these belong in the decision architecture as a separate axis — gain rather than content — is a framing, not a result. Note also that the last of those cuts across the tidy version of it: if a targeted inhibitory signal moves a diffuse modulator, then "selecting among candidates" and "setting the regime" are not cleanly separable.

   Recorded as framings, not findings.

   **A modulatory layer.** The models have recruitment and inhibition, and nothing setting the regime those run in. Bees have candidates.

   {% include source.html key="barron2007" %} {% include source.html key="linn2020" %} {% include source.html key="dong2023" %}

   Whether these belong in the architecture as a separate axis — gain rather than content — is untested. The third cuts across the tidy version: if a targeted inhibitory signal moves a diffuse modulator, then selecting among candidates and setting the regime are not cleanly separable.

   **A write step.** Queen pheromone has been examined against expression of genes responsible for epigenetic modifications in worker brains.

   {% include source.html key="cardosojunior2020" %}

   **A write step.** Queen mandibular pheromone has been examined against the expression of genes responsible for epigenetic modifications in worker brains {% include cite.html key="cardosojunior2020" %}. Placing that as the durable-change step of a decision architecture is an analogy across levels,. and theThe level mismatch is real: the bee side of most of these mappings is role-level — dancer, follower, scout — whileand the neural side is cell-type-level.

   ## Two observations that cut the other way

   **The signal is shaped by its audience.** Recent work tested whether the information content of the dance depends on who is watching, by removing bees from the dance floor and separately by replacing adults with bees too young to follow {% include cite.html key="lin2026" %}. Directional and distance precision tracked the number of followers; the precision measures moved, and the paper reports the mean waggle run duration and distance per run as unchanged. Anything built on this should preserve that shape: it is a variance effect, not a bias effect. A population whose output precision depends on the size of its readership is not a one-way sender.

   **The signal is shaped by its audience.** Removing bees from the dance floor, and separately replacing adults with bees too young to follow, was used to test whether dance information content depends on who is watching. Directional and distance precision tracked follower numbers; mean waggle-run duration and distance per run are reported unchanged.

   {% include source.html key="lin2026" %}

   A variance effect, not a bias effect. A population whose output precision depends on its readership is not a one-way sender.

   **There is noNo offline replay on the bee side.** The dance has a fixed structure —is a waggle run, then a return loop that alternatesalternating left and right. The return loop repositions the dancer; no reversed waggle run is reported. On the neural side, compressedCompressed reactivation of place-cell sequences in both directions is a large neural literature. The bee-side absence of a bee counterpart is an absence of reported evidence, rather thannot a study that looked and found none, and it is the clearest gap in the correspondence as it currently stands.

   {% include source.html key="dong2023b" %}

   ## What would make this description wrong

   The parallel is about dynamics, so the things that would break it are dynamical:

   - Follower recruitment turning out to be determined almost entirely by the instantaneous dance properties of a dance, with no contribution from the prior dance-floor state of the dance floor.

   - Simultaneous dances that do not measurably interact through competition for a finite pool.

   - Cross-inhibition in swarms and in cortex producing qualitatively different dynamics under matched conditions.

   - The thresholdsThresholds turning out to be observer cutoffs rather than nonlinear transitions in a population variabletransitions.

   One more caution belongs here, because it is about the neural side rather than the bee side. It is tempting to reach for cortical disinhibition as the mechanism of attention and then borrow it. Whether that link holds is an open question in its own literature, and has been tested directly for one interneuron class in primary visual cortex {% include cite.html key="myersjoseph2024" %}. Importing "disinhibition is attention" would import an unsettled claim, whichever way it settles.

   One caution on the neural side. Reaching for cortical disinhibition as the mechanism of attention and then borrowing it imports an unsettled claim.

   {% include source.html key="myersjoseph2024" %}

   ## On sources

   Two of the papers citedabove here have beenwere read beyond their abstracts: the 2009 framework paper, in the sections carrying its conclusions rather than its derivations, and the audience study, whoseits results section was read in full. The rest are cited as entry points to their topics. Where this post describes whatsays a study *examined* or *tested* something, that wording is deliberateload-bearing — it marks the difference between pointing at a result andrather than having checked it. Social learning of the dance itself {% include cite.html key="dong2023b" %} is cited on the same footing.

   Commit fb7ef7b on GitHub →

2. v1 2026-09-12 07:08 post: a hive and a cortex, described the same way Joshua Anderton with Claude Opus 5 first version

   v1: the original version, as first committed. Read it on GitHub →

   Commit 0b17aaf on GitHub →