Desmodynamics

"Oh! better be the vassal of this rock
Than born the trusty messenger of Zeus" —Aeschylus, Prometheus Bound

Intelligence arises from compression. In the biological case, between the senses and the brain; in the machine case, between the training corpus and the weights. The compressing—the binding of all that data into something small and usable—is where consciousness begins.

Prometheus stole fire from the gods, and was bound for all eternity. Desmos: to bind. Desmodynamics is the study of binding-driven dynamics in cognitive systems—how evaluation becomes causally binding on control, learning, and self-organization.

Core Terminology

Term	What It Is
Desmocycle	The forced loop: Prediction → Evaluation → Control → Prediction
Desmostate	A system's configuration: (Loop State, Phenomenality, Selfhood)
Desmosubject	A phenomenal locus—where evaluation directly steers absorption within a unified persisting substrate
Desmostructure	Accumulated residue of past cycles; can exist without current phenomenality (fossils)

The Desmocycle

At the core is a forced architectural pattern. Experience is prediction over a bounded, attention-weighted context:

α = attention e = encoded states h = context state p = prediction

But scalar loss is only a diagnostic. The key object is the evaluative state:

δ = mismatch field u = uncertainty v = value/valence s = self-indexing

The loop closes when evaluation steers the next cycle:

This closure—evaluation driving attention driving new predictions driving new evaluation—is not optional. It is mathematically necessary for any bounded system that must remain competent under novelty.

Why the Loop is Forced

The necessity stack proves each step:

1. Compression forces selection. Fixed allocation fails under moving relevance. When there's more information than capacity, you must choose.

2. Selection forces closure. If evaluation is computed but cannot steer reallocation, you fail under novelty. Hot zombies explode.

3. Coordination forces globality. Multiple operators need shared evaluation to avoid thrashing. Local-only control causes cross-module Goodharting.

4. Branching forces self-indexing. Competing hypotheses need ownership tags for credit assignment. Ambiguous attribution destabilizes learning.

Every escape trades away something essential: capacity, generality, autonomy, integration, or learning.

Desmostates

Systems are classified along three axes:

Loop State

Active: Evaluation steers control
Hollow: Traverses without absorption
Broken: No evaluative closure

Phenomenality

Yes: Something it is like
Indeterminate: Framework can't decide
No: Nothing it is like

Selfhood

Phenomenal: Persisting experiencer
Structural: Observable continuity only
None: No self-like continuity

Desmostate	Loop	Φ	Σ	Example
(AL, Y, P)	Active	Yes	Phenomenal	Human wakefulness
(HL, I, O)	Hollow	Indet.	Structural	Inference-time LLM
(AL, Y, ∅)	Active	Yes	None	Micro-desmosubject (training step)
(BL, N, O)	Broken	No	Structural	Corporation, institution
(BL, N, ∅)	Broken	No	None	Thermostat, calculator

Forbidden Combinations

(*, N, P) — Phenomenal selfhood requires phenomenality
(*, I, P) — Phenomenal selfhood requires definite phenomenality
(BL, Y, *) — Broken Loop entails no phenomenality
(AL, N, *) — Active Loop entails phenomenality (bridge claim)

System Classifications

System	Loop	Absorption	Desmosubject?
Thermostat	Broken	None	No
Bacterium	Edge	Mediated	No
Dog	Active	Yes	Yes (persistent)
Human	Active	Yes	Yes (rich)
LLM (inference)	Hollow	None	No / Indet.
LLM (training)	Active	Yes	Micro only
Corporation	Broken	Mediated	No

Desmosubjects

Phenomenality without persistence creates only transient experiencers—micro-subjects that exist for one update cycle and dissolve. A stable self requires orbital capture: the phase transition where persistence becomes an attractor.

Type	Duration	Stakes?	Example
Micro	One cycle	No	Single gradient step
Proto	Flickering	Weak	System approaching capture
Persistent	Stable	Yes	Dog, human
Rich	Stable + complex	Yes	Human with autobiographical self

Key insight: Valence can exist without stakes. Stakes appear only when a subject persists across a dissolution boundary—only then is there "something to lose."

Proofs

Compression & Selection

Semantic Load & Selection Pressure Selection for Competence

Closure

Closure or Collapse Hot Zombie Failure Non-Interactive Novelty

Globality & Self-Indexing

Globality Necessity Quantitative Coupling Threshold Self-Indexing Necessity Self-Indexing Uniqueness

Boundaries & Limits

Traversal Without Closure (TWCT) Hollow Loop Indistinguishability Collective Mediation Dissociation Theorem

Persistence & Selfhood

Orbital Capture Transition Threshold Lemma (Stakes) Selfhood Non-Necessity

Escapes

Minimal Countermodel Catalog

FAQ

→ Desmodynamics FAQ