The bat — active inference you can measure
How to read this page
A Plain and a Clear version of this page have not been written yet. What follows is the document itself.
Precise — the source document
This is the document. Rendered from the repository at the commit above, with nothing rewritten for the web. A gate re-renders it on every deploy and fails the build if a single byte differs.
RENDERED FROM STRUCTURED DATA — this page is not a markdown file. Source:
reader/plates/PL-07.jsoninuni-cookbookat575fc93d9d31. The sha256 recorded for this page is the digest of that JSON, not of this text. The renderer isgenerators/derive_docs.cjs; it copies fields and adds no facts. Every value below is a field of the source, verbatim, including the ones that readNOT RUNandPENDING.
The drawing below is
reader/plates/PL-07.svgat sha256f2a671c38cc053a6.
Six numbered readings of bat echolocation as a measurable free-energy loop, drawn from CN-10. PANEL I draws the sonar loop as one turn of a cycle — emit, echo returned at 2R = ct, belief update, call adaptation — and marks the echo-to-belief-update arrow with the double arrow as THE EPISTEMIC ARROW, with the type distinction printed on the plate: that arrow carries the REALIZED update D_KL[q(s|o,pi) || q(s|pi)] in nats after the echo has arrived, whereas the epistemic term of G(pi) is that update's EXPECTATION under q(o|pi), taken at the emit node BEFORE emitting. The arrow reports an outcome; the term scores a policy. The call costs joules and dB and returns an echo that does not feed, warm or move the bat: emission cost paid for information return, both sides on a meter. PANEL II draws range as delay, 5.83 ms per metre, on a log ruler. PANEL III draws the ~30x collapse of call duration across one attack and the overlap-free floor R >= c*tau/2 collapsing with it. PANEL IV draws the terminal-buzz rate rise beyond 160 calls/s and records the refuted overlap explanation (1.07 m) first-class: laryngeal motor performance binds first. PANEL V draws Doppler-shift compensation to scale — an 83.0 kHz emission returns at 85.4 kHz when closing at 5 m/s, roughly 1 kHz clear of the fovea's 84.5 kHz upper edge, and the bat lowers its emission ~2.4 kHz to 80.6 kHz to land the echo back on F_ref — and types it as an AMBIGUITY-minimising action, NOT the epistemic term and NOT policy precision gamma. PANEL VI draws the price of the epistemic term as a knee at ~130 dB SPL @ 0.1 m, with NO y axis, because the corpus carries no cost value. PANEL VII carries the key, this caption, and what the plate does not claim. Every number carries units, a ledger row and an evidence class. Zero UNI claims.
Claims
| ref | symbol | value | units | scope |
|---|---|---|---|---|
| c | ~343 | m/s | air, 20 C, 1 atm | |
| t(R) | 5.83 | ms per metre of range | t = 2R/c, two-way | |
| tau_search | 15-20 | ms | search-phase call duration | |
| tau_approach | 2-5 | ms | approach phase | |
| tau_buzz | 0.5-1 | ms | terminal buzz | |
| R_min | 2.6-3.4 / 0.34-0.86 / 0.086-0.17 | m (search / approach / buzz) | overlap-free floor R >= c*tau/2 | |
| rate_buzz | >160 (up to ~170) | calls/s | terminal buzz repetition rate | |
| f_muscle | up to 160; 200 in one case | Hz | anterior cricothyroid, Myotis daubentonii, positive work in cyclic contraction | |
| buzz ceiling cause | laryngeal motor performance, NOT pulse-echo overlap | - | M. daubentonii | |
| R_unamb@160/s | 1.07 | m | c*PI/2, PI = 6.25 ms | |
| f_fovea | 83.0-84.5 | kHz | R. ferrumequinum inferior colliculus, overrepresented best frequencies | |
| SD_echo | 110 (= 0.17% of F_ref) | Hz | Hipposideros armiger (a hipposiderid, NOT a rhinolophid), in-flight DSC precision. An emission-control statistic, NOT a measured resolution requirement of the receiver | |
| band_DSC | 0.1-0.2% of F_ref (~83-166 Hz at F_ref = 83 kHz) | % of F_ref | rhinolophids and P. parnellii - the BAND: precision within which F_echo is held around F_ref | |
| offset_DSC | ~150-200 | Hz (F_ref above F_rest) | R. ferrumequinum, R. euryale, P. parnellii, in flight - the OFFSET, not the band | |
| df_emit | ~2.4 | kHz (lowering) | bat at v = 5 m/s, f_r = f_e(c+v)/(c-v), F_ref = 83 kHz | |
| fovea width vs DSC precision, same animal | - | - | within-species pairing | |
| bat range-discrimination threshold | - | - | measured psychophysics (the 'jitter' literature) | |
| SL_open | ~130, up to and beyond 140 | dB SPL re 20 uPa @ 0.1 m | open-space aerial-hawking bats | |
| SL_whisper | up to 110 (not ~70) | dB SPL re 20 uPa @ 0.1 m | 'whispering' bats | |
| p@140dB | 200 | Pa (~0.2% of 1 atm) | p = 20e-6 x 10^(SL/20) at 0.1 m | |
| AGC | ~6 | dB sensitivity drop per halving of target distance | approach phase, attributed to middle-ear muscles | |
| cost of echolocation in flight | negligible at low intensity; exorbitant above ~130 dB SPL @ 0.1 m for small bats | - | Rhogeessa io (5 g); Pipistrellus nathusii | |
| rate_pulse,flight | 19.7 +/- 2.7 (range 15.3-25.8) | pulses/s | Rhogeessa io, in flight (non-buzz) | |
| G(pi) | = risk + ambiguity = -epistemic - pragmatic | nats | discrete POMDP, convention q(o,s|pi) = p(o|s) q(s|pi); the epistemic form further assumes q(s|o,pi) ~= p(s|o) | |
| biological systems observed to explicitly compute G(pi) | - | count | any taxon | |
| log shift, delay ruler vs range ruler | 0.766 | decade | t = 5.83*R => log t = log R + log 5.83; on a log ruler the delay scale IS the range scale, rigidly shifted | |
| R_min = c*tau/2 as a log shift | log10 2 (~0.30) | decade | on a log ruler the overlap-free floor is the c*tau bar shifted left by a constant, the same for every phase | |
| 110 Hz as a fraction of 83 kHz | 0.13 | % of 83 kHz | CN10-24 states 110 Hz = 0.17% of H. armiger's F_ref. 0.17% != 0.13%, therefore H. armiger's F_ref is NOT the 83 kHz drawn on this plate's axis, and the two must not be multiplied together. This is the arithmetic that exposes the CN10-30 splice |
What this plate does NOT claim
- That a bat computes expected free energy. Nothing on this plate shows a bat evaluating G(pi) over a policy space, or anything isomorphic to it. G(pi) in nats has never been decoded from any taxon (NOT-MEASURED, NA-02). 'Minimising the ambiguity term' is OUR description of its behaviour, in our vocabulary, for our purposes. The bat is not doing our arithmetic. (CN-10, Not claimed.)
- That Doppler-shift compensation is gamma (policy precision). gamma is the softmax inverse temperature over policies, units nats^-1. The fovea is SENSORY precision - the sharpness of p(o|s). Different objects, different units. DSC lowers the AMBIGUITY term of G(pi). Conflating them is a category error and CN-10 refuses it.
- That the epistemic term and the ambiguity term are the same term. G = risk + ambiguity and G = -epistemic - pragmatic are two decompositions of one number, and the terms do not map one-to-one: ambiguity != -epistemic, risk != -pragmatic.
- That the active-inference reading is required to explain any measurement here. Classical sensorimotor control, signal-detection theory and plain optimal-foraging accounts predict much of it. No discriminating experiment is offered, and none is known to the corpus in this pass.
- A within-animal pairing of fovea width and DSC precision. NOT-MEASURED (CN10-30): R. ferrumequinum's fovea against H. armiger's DSC is a splice across two species and two families. The chapter's '~14x finer than the fovea is wide' is illustrative, not a measurement. Panel V prints the arithmetic that exposes the splice and does not draw SD_echo on the 83 kHz axis.
- Any per-phase call-rate ladder. The corpus carries exactly two rates: 19.7 +/- 2.7 pulses/s non-buzz (R. io, Voigt & Lewanzik 2012) and >160 calls/s buzz (Elemans et al. 2011; Moss & Surlykke 2010). Different species, different studies. Panel IV draws two levels and refuses to draw a curve.
- Any cost of echolocation in J, W or ml O2. Panel VI has no y axis because the corpus has no value to put on one. The knee's LOCATION (~130 dB SPL @ 0.1 m) is sourced; the curve's SHAPE and HEIGHT are not.
- Any bat range-discrimination threshold. The 'jitter' psychophysics is contested and NOT-SOURCED (CN10-32). No threshold is implied by any tick spacing on Panel II.
- That the terminal-buzz ceiling is set by pulse-echo overlap. Refuted (CN10-21). The 1.07 m is drawn precisely because it is wrong: a derivation that predicts the right number for the wrong reason.
- That a bat is conscious, aware, or has experience. Nagel's question is not answered, addressed, or bounded by any measurement on this plate. A permanent OPEN QUESTION, in a different ledger. 'Full human' and 'beyond human' appear here as no target, milestone or deliverable.
- That any citation on this plate raises any UNI rung. A NATURE CITATION IS NEVER A UNI GATE. The NATURA classes and the UNI 4-value fence (proven / designed / hypothesized / not-yet-built) describe different kinds of claim and never merge. ZERO UNI CLAIMS ON THIS PLATE. Program position, unsoftened: ~2 of 11+ rungs earned; a developmental active-inference SIMULATION; a toy world, never a person.
- That any bat trait drawn here is an optimum. Per Gould & Lewontin (1979), phylogenetic inertia, drift, developmental constraint and frozen accidents produce traits that solve nothing. Nature's authority here is precise and limited: it already ran the search under real physical constraints and deleted the failures.
Source chapters
cookbook/recipes-natura/CN-10-bats.mdencyclopedia/wing-NATURA/NA-02-the-one-loop.mdencyclopedia/NATURE-LEDGER.md
sha256 29d6413b7c704115 — of the original file, so what was ingested stays checkable.