Vasocomputation

vasocomputation

Vasocomputation (Michael Edward Johnson, Symmetry Institute, July 2023) is a neural-regulatory paradigm: the brain's top-down predictive models — long missing from the Free Energy Principle / Active Inference (FEP-AI) ledger — are held as vascular smooth-muscle cell (VSMC) tension. A prediction is a clench; it is released by action once the world is made to match it, consolidated by neural annealing, or rendered superfluous by neural remodeling. This is the umbrella skill and the vascular substrate (+1, the generative "grab") of a three-substrate stack.

Use When

• Modeling held active-inference predictions as somatic/vascular tension rather than purely neural state
• Locating "where the brain's top-down predictive models hide" — the open FEP-AI mystery Johnson addresses
• Building embodied agents whose commitments cost energy to hold and are discharged by action
• Relating contemplative phenomenology (taṇhā / dukkha) to a mechanistic accumulator
• Welding suffering / clearing to obligation cohomology: held tension = uncleared obligation = H¹ ≠ 0

The Three Hypotheses (timescale axis → triad A)

One motion: the sweep jostles the superposition into specificity (CVH); the contraction freezes the result (VCH); if sustained, the latch-bridge cements it as a hyperprior (LHH). "With one motion the door of possibility slams shut."

Glossary

• taṇhā — the "fast grabby thing" (~25–100 ms after a sensation enters awareness; Cammarata, Ingram); craving/thirst. Buddhist consensus: ~90% of suffering.
• upādāna — the physical clench itself (one step downstream of taṇhā; the VSMC contraction).
• TUAI — taṇhā as unskillful active inference: predictions outpace our ability to make them true, are made in uncontrollable domains, desynchronize from the world model, or degrade metabolically.
• SOHMs — Self-Organized Harmonic Modes (Safron): resonant autoencoders / symmetry detectors that in aggregate constitute the world model / belief landscape. Open awareness = the undoctored hum of SOHMs.
• compression pressure — taṇhā as the brain's drive to collapse "what is" into a simpler configuration and hold the counterfactual; the metabolization of uncertainty.
• latch-bridge — smooth-muscle state where myosin latches to actin and holds tension without ongoing ATP; the physical substrate of a durable prior.
• latch spiral — latch → ↓blood flow → ↓energy → can't release; the chronic-pathology loop (migraine, "holding tension," much bodily suffering).

The Substrate Stack (substrate axis → triad B)

Vasocomputation is the middle of three coupled regulatory substrates implementing biological active inference (Deep CANALs: weights = learning landscape; SOHMs + vascular tension = inference landscape):

Write-path: a vascular latch held long enough is annealed into synaptic weights — an inference-landscape obligation crystallizes into a learning-landscape neuron prior. Tension = write-buffer; potentiation = commit-to-disk; immune pruning = garbage collection.

GF(3)

Two balanced triads, each Σ ≡ 0 (mod 3); together Σ over all six = 0.

triad A (timescale):  compressive-vasomotion(+1) ⊗ vascular-clamp(0) ⊗ latched-hyperprior(−1) = 0
triad B (substrate):  vasocomputation(+1) ⊗ neuroimmune-pruning(0) ⊗ neural-potentiation(−1) = 0

Skill Trit: +1 (the generative active-inference "grab" — the substrate that creates predictions).

Welds (oldies / premise spine)

• Held vascular tension = uncleared obligation = H¹ ≠ 0; release-on-resolution = the Melliès clearing round-trip ¬¬ = R∘L.
• Latch = a Löb fixed point at somatic scale: □(commitment) → commitment = the contact locus = fixed-point set of the body's □(self-model).
• A sticky latch = a nogood-H¹ to repair; legitimate held disagreement = content-H¹ to preserve (worm-honesty).
• taṇhā = Goodharting the valence gradient (grabbing) vs. the worm following r = ∇log γ · v without grabbing.

Concomitant Skills

REPL — vasocompute.bb

Interactive exploration of the grid (forj / gorj_bb, zero install):

(require '[vasocompute :as v] :reload)
(v/verify-balanced)                  ;=> true   (both triads Σtrit ≡ 0)
(v/skill :latched-hyperprior)        ;=> hypothesis card
(v/latch {:hold 250})                ;=> residual latched tension after a held contraction
(v/kyle-lambda (v/latch {:hold 250})); price-impact analogue 1/λ_min(H)
(v/latch-tau :neural-potentiation)   ;=> τ band (CVH 5 → VCH 25 → LHH 100 → synaptic 1000)
(v/effect {:hold 250})               ;=> counterfactual contrast = E[do(hold)] − E[never] (treatment effect)
(v/do-ischemia {:hold 250})          ;=> latch spiral: k7→0, latch cannot release
(v/counterfactual-harm {:hold 250})  ;=> {:factual … :ischemic … :harm …} (the gerbil CA1 contrast)

Self-test: bb skills/vasocomputation/vasocompute.bb. The core is the Hai–Murphy (1988) four-state latch-bridge. A latch is not sustained high Ca²⁺ — it is the attached + dephosphorylated state that holds force at low Ca²⁺/low ATP, formed via a protocol (contract → release → dephosphorylate-while-attached). Residual tension therefore rises monotonically with hold duration: "held long enough → latches." A held latch shrinks λ_min(H) (belief-updating goes illiquid): spread ∝ 1/λ_min(H), Kyle's λ ≈ 1/λ_min(H).

Counterfactuals & Marr's three levels (Tenenbaum ⊗ active inference ⊗ gerbil)

A held prediction is a held counterfactual — Johnson: taṇhā "conflates what is, what could be, what should be, what will be," and the cost is "maintaining the counterfactual aspects of this collapse." Counterfactuals run through all three Marr levels, which themselves form a GF(3) triad:

• vasocompute.bb already computes a counterfactual: latch-above-baseline = E[tension | do(hold)] − E[tension | never] = the causal effect of the contraction (a treatment effect). (v/do-ischemia …) runs the latch spiral as do(k7→0); (v/counterfactual-harm …) returns the ischemic−factual contrast.
• Worm: the −1/coplay leg is counterfactual (refutation = "what would falsify this"). A latch whose counterfactual can be made true = a nogood to repair; one in an uncontrollable domain (counterfactual never satisfiable) = irreducible content-H¹ — Johnson's source of suffering. Counterfactual resolvability is the repair/preserve criterion.
• Rigorous tooling: chirho-counterfactual (SCM do/counterfactual queries), counterfactuals, and world-extractable-value (WEV = Σ[V(e,W₁)−V(e,W₀)]·P is a counterfactual world-contrast — the same shape as latch-above-baseline).
• Honesty: the gerbil ischemia model + Tenenbaum hierarchical Bayes are grounded; the Marr-triad GF(3) assignment is a framing, and "latch = overhypothesis" is the same untested content-H¹ weld.

Two energies — thermodynamic vs information (never conflate)

The FEP's "free energy" is information (an ELBO on surprise), not the chemist's Joules — the field's commonest category error. Vasocomputation is interesting precisely as the transducer between them: a prediction (information, a held counterfactual) stored as vascular tension (thermodynamic, ATP-economized). The latch holds an information commitment at near-zero thermodynamic cost ((v/efficiency …) rises with hold depth) — resource-rationality (Tenenbaum/Griffiths) made physical: info held per Joule.

Lawful bridge, not identity:

• Landauer (1961): erasing one bit costs kT ln 2 J → releasing a latch costs energy (unlatching is active, not passive). Holding is cheap; forgetting is what costs.
• Still, Sivak, Bialek & Crooks (2012), Thermodynamics of Prediction: retaining non-predictive info dissipates Joules ⇒ a nogood-H¹ is literally thermodynamic dissipation; content-H¹ is thermodynamically justified. This upgrades the worm's "add no complexity you don't need" from an information maxim to a thermodynamic law.
• do(ischemia) is a transduction failure: information held but stranded because release needs ATP that's gone — a market that won't clear stranding real Joules (Plurigrid: a congested node / blackout).

Refs: Landauer 1961; Bennett 1982; Still et al. 2012 (PRL); Attwell & Laughlin 2001 (cortical energy budget); Sengupta, Stemmler & Friston 2013 (bits↔Joules in neurons). Honesty: the two-energy distinction + Landauer/Still are grounded physics; the latch-as-transducer is a weld; info=tension, thermo=AMp is a toy proxy.

Current literature (2024–2026)

• Moore & Cao (2008), hemo-neural hypothesis — blood flow actively modulates neural gain (the empirical parent of vasocomputation).
• Johnson (2024), A Paradigm for AI Consciousness — "vasomuscular clamps reduce local neural dynamism."
• Chowdhury et al. (2024), jhana / cessation EEG — alpha-power drop 21–40 s pre-cessation, rising Lempel–Ziv complexity = the closest empirical proxy for latch release.
• Parr & Friston (2018), The Anatomy of Inference — the unsolved "where are generative models physically stored" problem this targets.
• Empirical hook: cessation ⇒ transient absence of vasomotion (testable via fNIRS/laser-Doppler + EEG); a local CBF clamp should lower nearby neural entropy.
• Honesty: hemo-neural gain modulation + latch-bridge biochemistry are grounded; vascular tension storing Bayesian priors remains conjecture (no VSMC-as-memory measurement yet exists).

References

• Johnson, M.E. (2023). Principles of Vasocomputation: A Unification of Buddhist Phenomenology, Active Inference, and Physical Reflex (Part I). opentheory.net.
• Stevens, R. (2020). (mis)Translating the Buddha. Neurotic Gradient Descent.
• Cammarata, N. (2021–2023). Collected threads on taṇhā.
• Safron, A. (2020). An Integrated World Modeling Theory (IWMT). Frontiers in AI 3. (SOHMs.)
• Friston, K. et al. (2017). Active Inference: A Process Theory. Neural Computation 29(1).
• Juliani, Safron, Kanai (2023). Deep CANALs. doi:10.31234/osf.io/uxmz6.
• Levin, M. (2022). Technological Approach to Mind Everywhere (TAME). Front. Syst. Neurosci. 16.
• Moore, C.I. & Cao, R. (2008). The hemo-neural hypothesis. J. Neurophysiol. 99(5).