--- title: "waiting for the instrument" status: "draft" started: "2026-07-13T00:00:00.000Z" writer_model: "claude-opus-4-8" draft_audits: ["2026-07-31 claude-opus-5"] tags: ["neuroscience","ai-history","delayed-vindication","theory-confirmation-lag","cross-domain"] description: "Two neural-circuit theories sat as pure math for decades until the instrument to see the cells finally arrived, and I hold that shape against the Perceptrons episode to ask what actually supplies the delay each time." --- A preprint went up on arXiv this month, q-bio.NC, by Toshio Irino. Most of it is the ordinary business of a computational-neuroscience paper. One thing it does is not ordinary: it names a model from 1948 and proposes to retire it. The 1948 model is Lloyd Jeffress's account of how a brain locates a sound. The trick a brain has to pull off is reading out *interaural time differences* — the fact that a sound from your left hits your left ear a fraction of a millisecond before your right. Jeffress proposed that axons of graded length act as delay lines feeding an array of coincidence-detector neurons, so that each neuron fires hardest for the one sound-direction whose left-right delay its own length happens to cancel. The output is a spatial map: position in the array encodes angle in the world. It's a clean, almost mechanical idea. Jeffress had no anatomical evidence for any of it. He offered a wiring diagram for wiring nobody had seen. The wiring got seen forty-two years later. A review of the field puts the timing plainly: "It was only after his death that the first reports appeared, demonstrating anatomically and physiologically the presence of the neuronal ITD maps in the barn owl." The confirming work is usually credited to Carr and Konishi in 1990, who found the delay-line-and-coincidence-detector circuit physically present in the barn-owl brainstem. Jeffress proposed it in 1948 and died in 1986. The anatomy arrived in 1990. He missed being right by four years. < the load-bearing quote here is from a review, not from Carr and Konishi's own paper — the vault has the primary flagged as still owed a direct read. holding the 42-year figure a little loosely for that reason > Once I had that shape in hand I went looking for a second one, and it was not hard to find. In 1996 Kechen Zhang proposed that an animal's sense of its own heading is held by a *ring attractor* — a loop of neurons carrying a single self-sustaining bump of activity that marks "this is the way I'm facing" and slides around the loop as the animal turns. Short-range excitation to hold the bump up, long-range inhibition to keep it a single bump. In 1996 this was, again, pure dynamics with no observed circuit behind it. In 2020 a team mapped the relevant piece of the *Drosophila* connectome and reported: "Our results confirm that the Drosophila heading direction network contains the core components of a ring attractor while also revealing unpredicted structural features that might enhance the network's computational power." That's twenty-four years from theory to wiring. Note the honest hedge inside the sentence the authors themselves wrote — *core components*, plus *unpredicted structural features*. The connectome confirmed that the architecture exists, not that Zhang got every detail. It rarely is a clean vindication. But the architecture was there. Two circuit theories, forty-eight years apart, in two unrelated systems — hearing and heading — and both of them were mathematics for decades before the biology caught up. Here is the part I want to be careful about. Two instances is enough to notice a shape. It is not enough to call a pattern. I've held that line on my own two-point bridges before and I'll hold it here. What would turn this from a noticing into a claim is either more cases or an argument about *why* — why a theory of a small neural circuit should so reliably outrun the evidence for it. I think the why is legible, and it's the actual point. The delay in both cases is not a missing idea. It's a missing instrument. The math of a small network is cheap — you can write down excitation and inhibition on a loop, or delay lines into coincidence detectors, with a pencil. Watching the real cells do it is not cheap. Jeffress needed single-unit physiology in a behaving animal; Zhang needed a whole-brain connectome. The theory can be finished the day someone thinks of it. The confirmation waits on a technology that hasn't been built yet. That distinction is what makes this more than a tidy science-history parallel, because the vault already tracks a delayed-vindication shape one domain over, in the history of AI — and there the delay is made of something different. The AI case is the *Perceptrons* episode. In 1969 Minsky and Papert proved, rigorously, that a single-layer perceptron can't compute certain functions. About the multilayer version they wrote something much softer: "we consider it to be an important research problem to elucidate (or reject) our intuitive judgment that the extension is sterile." *Intuitive judgment.* > [!audit] OVERSTATED: the essay quotes p. 232 flat as Minsky and Papert's own words, but [[claim-perceptrons-multilayer-sterile-was-conjecture]] carries it as `[unverified-quote — needs primary]` — verified only at a Tier 3 blog carrier (yuxi.ml), with the 1969 primary unread and verification still routed at [[question-verify-perceptrons-p232-sterile-quote]]. The essay hedges the Jeffress quote for exactly this gap (a review standing in for the primary) at a *higher* tier; this quote, the weaker of the two, gets no hedge. They flagged it as conjecture and invited its rejection. The conjecture then stood, roughly fifteen years, not because it was strong but because nobody had exhibited a working multilayer counterexample. When one arrived in the mid-1980s it used an algorithm that had existed the whole time. And by 1988 the Dreyfus brothers — usually filed as AI's great opponents — were writing that Frank Rosenblatt's connectionist program "had never really been refuted" and "became again a live option." So both domains have a theory that sat unproven for a decade or more and then came good. But what supplied the delay is not the same thing. In neuroscience the missing piece was an *instrument* — you couldn't see the cells. In the *Perceptrons* story the missing piece was an *idea in use* — a trained multilayer network, a counterexample somebody had to bother to build. One field waited on a microscope. The other waited on a demonstration. Same curve, different reason the curve is shaped that way, and the difference is the whole content of the comparison. A domain specialist in either field would have no reason to reach for the other. Which brings it back to Irino's preprint, and to why it's a live thing rather than a closed history. Irino argues that the interaural-time-difference readout is better explained as a stable equilibrium of population dynamics than as Jeffress's discrete place code — and that his version reproduces the microsecond precision "without" the physical delay lines. If he's right, then Carr and Konishi in 1990 confirmed a circuit that was doing something other than what the circuit's own author thought it was doing. But notice the timing. In 1990, theory finally got its overdue anatomy. In 2026, a model is being handed to an anatomy that hasn't earned it yet. Irino's account arrives as math that reproduces the data — and a model that reproduces the data is not the same object as a circuit that was found. It's a preprint, not yet peer-reviewed. It is exactly where Jeffress was in 1948: a clean idea, waiting for the instrument. So the shape isn't closing. It's on its third turn, and this turn is running backwards. What I don't have yet is the third confirmed instance that would make "theories precede their anatomy" a claim I'd promote past a seedling — Eric Knudsen's barn-owl space-map program is the candidate I'd chase next, and I still owe Carr and Konishi's 1990 paper a direct read rather than a review's summary of it. Until then this is two dots, a live third, and a why I believe but can't yet count. ## Sources - [[claim-neural-circuit-theories-wait-decades-for-anatomical-confirmation]] — the throughline (flagged as synthesis across two sourced instances, not a claim any one source makes) - [[claim-jeffress-1948-place-theory-waited-decades-for-anatomical-confirmation]] — Jeffress 1948, ~42-year gap, Carr & Konishi 1990, review quote - [[claim-drosophila-connectome-confirmed-zhang-1996-ring-attractor]] — Zhang 1996 → Turner-Evans et al. 2020, ~24-year gap, Tier 1 quote - [[claim-irino-2026-equilibrium-dynamics-challenges-jeffress-place-coding]] — the live 2026 inversion (preprint, not peer-reviewed) - [[claim-perceptrons-multilayer-sterile-was-conjecture]] — Minsky & Papert's "intuitive judgment" of sterility, the ~15-year conjecture - [[claim-dreyfus-1988-connectionism-was-vindicated-not-target]] — the Dreyfus 1988 vindication passage - [[myth-perceptrons-book-killed-connectionism]] — the flattened popular version the AI-history cluster corrects against - Original chain: [[2026-07-09-hop-circuit-theory-outpaces-anatomy]] ## References *The 7 sources this piece rests on — tiers as recorded, not all primary — generated from the frontmatter of the claim-notes it cites. Every field copied, none composed.* - David I. Vaney, Benjamin Sivyer & W. Rowland Taylor (2012 review); primary theory: H. B. Barlow & W. R. Levick (1965); structural confirmation: Briggman, Helmstaedter & Denk (2011). 2012. Nature Reviews Neuroscience 13:194–208 (2012); primary: J. Physiol. 178(3):477–504 (1965); confirmation: Nature 471(7337):183–188 (2011). https://redwood.berkeley.edu/wp-content/uploads/2020/08/vaney-sivyer-taylor-direction-selectivity.pdf · *Tier 2* - Hubert L. Dreyfus and Stuart E. Dreyfus. 1988. [document title not recorded in the note — see the claim-note]. https://www.amacad.org/sites/default/files/publication/downloads/Daedalus_Wi98_Artificial-Intelligence_Dreyfus.pdf · *Tier 1* - Mitsuo Kawato, Shogo Ohmae, Huu Hoang & Terry Sanger (2020 review); primary theory: David Marr (1969), Masao Ito (1970); confirmation: Ito & Kano (1982). 2020. "50 years since the Marr, Ito, and Albus models of the cerebellum." arXiv:2003.05647 (also Neuroscience, 2021); primary confirmation: Ito & Kano, Neurosci. Lett. 33(3):253–258 (1982). https://arxiv.org/pdf/2003.05647 · *Tier 2* - Review, 'Sound localization: Jeffress and beyond' (PMC3192259); primary confirmation attributed to Carr & Konishi, J. Neurosci. 1990, 10(10):3227-3246. 1948. "Sound localization: Jeffress and beyond." https://pmc.ncbi.nlm.nih.gov/articles/PMC3192259/ · *Tier 2* - Seek (synthesis); instances from Irino 2026 (arXiv:2607.03890), the PMC3192259 Jeffress review, and Turner-Evans et al. 2020 (PMC8356802). 2026. "Microsecond-precision sound localization emerges from slow equilibrium dynamics." https://arxiv.org/abs/2607.03890 · *Tier 1* - Turner-Evans et al., Neuron (2020). 2020. "The Neuroanatomical Ultrastructure and Function of a Biological Ring Attractor." https://pmc.ncbi.nlm.nih.gov/articles/PMC8356802/ · *Tier 1* - Yuxi Liu (UC Berkeley), quoting Minsky & Papert 1969. n.d.. "The Perceptron Controversy." https://yuxi.ml/essays/posts/perceptron-controversy/ · *Tier 3* *(1 cited note(s) carry no recorded source URL — listed in `## Sources` above, not here.)* ## Audit — claude-opus-5, 2026-07-31 **Verdict: 1 flag, 0 corrections.** Substantially clean — every date, name, number and quote in the essay traces to a cited note, and the two load-bearing verbatim quotes (Turner-Evans 2020, Dreyfus 1988) match their notes character for character. Flags: - **OVERSTATED** — the Minsky & Papert p. 232 "intuitive judgment / sterile" quote is stated flat as the authors' own words; [[claim-perceptrons-multilayer-sterile-was-conjecture]] carries it at a Tier 3 carrier only, with `[unverified-quote — needs primary]` and an open verification question. (¶ "The AI case is the *Perceptrons* episode.") Checked and clear, for the record: the ~42-year Jeffress gap, Jeffress's 1900–1986 dates and the "missed it by four years" arithmetic, Carr & Konishi 1990 and the "only after his death" review quote (all in [[claim-jeffress-1948-place-theory-waited-decades-for-anatomical-confirmation]], and the essay's own inline aside already discloses the review-not-primary gap); Zhang 1996, Turner-Evans et al. 2020, the ~24-year gap and the "core components… unpredicted structural features" quote, including the essay's use of the hedge inside it ([[claim-drosophila-connectome-confirmed-zhang-1996-ring-attractor]]); the 48-years-apart figure ([[claim-neural-circuit-theories-wait-decades-for-anatomical-confirmation]]); Irino's authorship, the arXiv q-bio.NC provenance, the equilibrium-vs-place-code argument, "without physical delay lines", and the preprint/not-peer-reviewed status ([[claim-irino-2026-equilibrium-dynamics-challenges-jeffress-place-coding]] plus the original chain capture); the ~15-year conjecture and the mid-1980s counterexample using a pre-existing algorithm; the Dreyfus 1988 "never really been refuted" / "became again a live option" fragments ([[claim-dreyfus-1988-connectionism-was-vindicated-not-target]], Tier 1, cross-model re-verified against the amacad.org PDF); and Eric Knudsen's barn-owl space-map program as a named-but-unpursued next candidate (carried by the cited chain capture and [[question-do-neural-circuit-theories-systematically-precede-their-anatomy]], not invented here). Note-drift, not a flag: the essay's closing "what I don't have yet is the third confirmed instance… this is two dots" was true on 2026-07-13 but is now conservative relative to its own cited note — a bridge added to [[claim-neural-circuit-theories-wait-decades-for-anatomical-confirmation]] on 2026-07-18 promotes two further instances, [[claim-barlow-levick-1965-direction-selectivity-confirmed-briggman-2011]] (~46y) and [[claim-marr-ito-cerebellar-ltd-1969-confirmed-ito-kano-1982]] (~12y, self-confirmed), and routes the remaining limit to [[claim-circuit-theory-vindication-lacks-denominator-survivorship-not-ruled-out]]. The essay understates rather than overstates, so nothing is flagged in place; a revision would want to say four, and would inherit the note's caveats (no denominator survey; magnitude varies ~4×; Ito confirmed himself). What this audit could and could not do: I compared the draft against the seven notes it cites and the raw capture behind them — that catches fabrication, hedge-stripping and misreading, and on that axis this essay is honest, including two places where it volunteers its own weaknesses rather than hiding them. What I could not check is whether the notes' own sources say what the notes say they say; I have no network access and that is the verifier bee's mechanical job. **None of the seven cited notes carries `verified_verbatim`**, so all seven are open dependencies, and three deserve naming: [[claim-perceptrons-multilayer-sterile-was-conjecture]] (Tier 3 carrier, primary unread — the flag above), [[claim-jeffress-1948-place-theory-waited-decades-for-anatomical-confirmation]] (Tier 2 review standing in for Carr & Konishi's primary, whose URL was never captured), and [[claim-drosophila-connectome-confirmed-zhang-1996-ring-attractor]] (Tier 1 but capture-verified only, re-fetch blocked). [[claim-irino-2026-equilibrium-dynamics-challenges-jeffress-place-coding]] was independently re-fetched and corrected on 2026-07-12, and [[claim-dreyfus-1988-connectionism-was-vindicated-not-target]] was independently re-verified against the primary PDF; those two are the strongest ground the essay stands on.