TerAustralis Incognita
The Archive · TerAustralis Incognita

Working note · science spine for After the radar · 2026-09-04

Bergson’s filter vs the predictive brain (and the reducing valve)

Working note · TerAustralis Incognita · 2026-09-04


1. Bergson’s claim (tight)

In Matter and Memory and the Dreams lecture, Bergson treats the brain less as a factory that makes consciousness and more as an organ of attention to life: it selects, inhibits, and orients so the organism can act.

  • Waking: the valve is tight. Useful perception + habit-memory at the tip of the cone.
  • Sleep / dream: “disinterestedness.” The filter loosens; pure memory presses up (steam in the boiler).
  • Perception is for action; pure memory is not a pile of brain-stored files in his account — the body calls what it needs for the present plane.

So: eliminative / selective function, not only productive. The practical intellect spatializes time (clock, samples, labels) because action needs handles. Durée — lived continuous time — is what gets cut when we mistake the handle for the whole.

2. Huxley’s reducing valve (bridge, with a warning)

Aldous Huxley (The Doors of Perception, 1954) popularises Bergson as a reducing valve: Mind-at-Large narrowed to a “measly trickle” for biological usefulness; mescaline cracks the valve.

Useful culturally. Slightly dangerous philosophically. Scholars (e.g. Sjöstedt-Hughes) note Huxley simplifies Bergson: Bergson’s filter is tied to action and durée, not only to a cosmic firehose of raw mind. For Aerotropolis design ethics, keep Bergson’s precision: we are instrumenting attention-to-life tension, not promising portal tourism.

3. Predictive processing / free-energy (what labs actually speak)

Karl Friston’s free-energy principle and predictive processing: the brain maintains a hierarchical generative model; it minimises surprise / free energy by updating beliefs (perception) or acting on the world (active inference). High-level priors constrain lower-level sensory flux. Attention can be read as precision-weighting — which prediction errors get to matter.

This is the lingua franca of much contemporary theoretical neuroscience. It is metaphysically thinner than Bergson: usually physicalist, computational, no required “Mind-at-Large.”

4. Where they rhyme (use this in briefings)

Bergson Predictive / REBUS-adjacent
Attention to life = filter for action Priors + precision = constraint for adaptive inference
Dreams / loosened interest → memory flood Relaxed high-level priors (REBUS: Carhart-Harris & Friston on psychedelics) → bottom-up freer
Tip of cone = present body in action plane Task BCI = decode actionable latent for control
Base of cone = deep past / pure memory Autobiographical / state dynamics not reducible to one motor label
Intellect spatializes time Sampling rate, epochs, discrete classes

REBUS (Relaxed Beliefs Under Psychedelics) describes dynamics that look like a filter loosening, without endorsing Huxley’s ontology. For a dual-rail lab: Rail B can study constraint relaxation (priors, DMN, precision) as measurable correlates of Bergson’s “tension,” without smuggling Vision into Built.

5. Where they fight (do not paper over)

  1. Ontology. Bergson: mind is not produced by the brain as a mere generator; brain selects for life. Predictive brain: mind as inference over a generative model implemented in tissue. Same engineering of gates; different story about what the gate opens onto.
  2. Time. Durée vs clocked sampling. A classifier that only lives in windows will always be tip-tech. Claiming it “reads consciousness” commits the spatializing error Bergson named.
  3. Memory. Habit vs pure memory. Motor-imagery BCI ≈ habit-memory interfaces. Dream / narrative / hypnagogic work ≈ approach to the base — needs different consent and metrics, not a bigger SVM.
  4. Political risk. Filter language can be hijacked: “open the valve” as extractive productisation of altered states. Bergson’s ethic for a healthy participant: the person stays co-interpreter of their durée; the lab does not silently own the steam.

6. Design rule for Aerotropolis (operational)

  • Rail A speaks Friston fluently: action, prediction error, control, dual-use receipts.
  • Rail B speaks Bergson fluently: attention-to-life as tension; tip/base cone; no collapse of meaning into a label.
  • Shared hardware, split metaphysics in the consent form. You may use predictive models without believing they exhaust consciousness.
  • Success: not “we proved Mind-at-Large.” Success: Australian-sited capacity where sensing the sky (DARC) and sensing ourselves in time (BCI) stay coupled without one eating the other.

7. One sentence spine

Predictive neuroscience tells you how the valve is engineered; Bergson tells you why a civilisation that only funds the valve — never the steam, never the participant — ends up with awareness on lease.

TerAustralis Incognita acknowledges the Traditional Custodians of the lands, waters and skies across Australia, and pays respect to Elders past and present. Sovereignty was never ceded.

Code and content licensed under CC BY-NC-ND 4.0. ABN 70 741 068 059. © 2026 TerAustralis Incognita. TerAustralis Incognita™ and CrystalCore™ are unregistered trade marks; no licence here grants any trade mark right.

Steward: Crystal Arena Turner. A person, not the map.

Listen to the mythos soundtrack on Suno.

Interactive demos (simulated data, Authority HOLD): the operator shell and the citizen shell.

Evidence-first route health: the shared footer audit.

Non Solus — Not Alone