assembly-economics/models/InformationOrganismsInGADM.md

<!--
Created 2026-10-12 under ASSEMBLY-WP-0003-T03 (INTENT.md next step 11, WS-13).
Sources: models/GenerativeAssemblyDynamicsModel.md (GADM-0.1),
concepts/AdaptiveInformationProcessingNetworkConcept.md, and the membrane,
seed-kernel, evolution, metabolism and resource-ontology sections of
history/20261011-InitialExploration.md.
-->

# Information Organisms in GADM — Mapping Note v0.1

## Purpose

GADM-0.1 models a society's assembly dynamics with ten stocks. It treats
automation only as part of Productive Capacity \(K\). The information-organism
lens describes machine capacity as populations of adaptive, governed, evolvable
systems: AIPNs with membranes, metabolism and lineages.

This note asks two questions:

1. Where do information organisms enter GADM-0.1's existing stocks and flows?
2. Which variables does GADM-0.1 lack once organisms are taken seriously?

Everything below is a hypothesis for GADM-0.2 and later, not a result.

---

## 1. Why organisms are not just more \(K\)

GADM-0.1 has \(\dot K = a_K V_{commercial} - \delta_K K\). Conventional
capital depreciates and is shaped from outside. Information organisms differ
in four ways that matter dynamically:

| Property | Conventional \(K\) | Information organisms |
|---|---|---|
| Change with use | depreciates | **learns**: may appreciate with use |
| Reproduction | built anew | **inherits and recombines** learned structure across generations |
| Boundary | none | **membrane**: what it senses and touches is governed and configurable |
| Inputs | capital, energy, labor | **metabolism**: compute, information, authority, and human attention |

**Hypothesis M1.** Folding organisms into \(K\) hides the two loops that
matter most for generativity: learning-with-use (a reinforcing loop inside
\(K\)) and lineage ownership (a reinforcing loop into \(P\)).

**Proposal.** Split productive capacity in GADM-0.2:

\[
K = K_{phys} + K_{\Omega}
\]

where \(K_{\Omega}\) is the capability of the information-organism population
\(\Omega\), with

\[
\dot K_{\Omega} = a_{\Omega} V_{commercial} + \ell \cdot U_{\Omega} \cdot Q - \delta_{\Omega} K_{\Omega}
\]

where:

- \(U_{\Omega}\) = utilization (organisms learn only when they run),
- \(\ell\) = learning yield per unit of utilization,
- \(Q\) multiplies learning: organisms learn from signals only as well as the
  observers and evidence around them allow.

---

## 2. Stock-by-stock mapping

| Stock | How organisms raise it | How organisms lower it | Lever from the organism lens |
|---|---|---|---|
| \(H\) Human Capability | tutoring, cognitive augmentation, freeing time for development | deskilling, loss of tacit knowledge when humans stop practicing | route-to-human actions, symbiosis rather than autonomy |
| \(X\) Assembly Commons | open lineages and inheritable learned structure become reusable components (\(I_X\), raises \(\rho\)) | proprietary lineages and closed membranes act as enclosure (\(L_X\)) | breeding regime: who may inherit from which lineage |
| \(K\) Productive Capacity | see § 1 | metabolic waste, over-provisioned compute | return-on-compute regulation |
| \(F\) Risk Capacity | automated underwriting, cheap simulation of experiments before real risk | correlated failures when many actors use the same lineage | diversity of lineages |
| \(G\) Effective Agency | personal or community-owned organisms raise *Access* and *Restartability* | *LockIn* through membranes controlled by others; delegation without exit | membrane portability, the right to own one's organism |
| \(N\) Diversity & Composability | recombination along interfaces; generalists and specialists coexisting | monoculture from one dominant lineage | interface standards for recombination |
| \(Q\) Epistemic Capacity | *Research*, *Audit*, *OpenEvidence* via observers, provenance and explainability | *Opacity*, *Manipulation*, *InformationCapture* | explainability and provenance as hard requirements |
| \(P\) Domination Concentration | — | *DiscoveryControl*, *InfrastructureControl*; ownership of lineages compounds | membrane interoperability, lineage ownership diffusion |
| \(E\) Ecological Headroom | efficiency gains, better ecological sensing | compute as energy draws directly on physical throughput | metabolic budgets tied to \(f(E)\) |
| \(S\) Social Cohesion | abundance that is broadly shared, fewer attention conflicts | attention extraction, loss of meaningful participation, opaque authority | attention protection; authority stays with humans |

---

## 3. Flow-level entry points

### 3.1 Experimentation \(V\)

Organisms lower the cost of serious experiments. This raises the baseline
\(v_0\), and partly decouples \(V\) from \(H\) and \(F\).

**Hypothesis M2.** Because GADM-0.1 notes that \(V\uparrow\) with
\(Q\downarrow\) accelerates damage, organism-driven experimentation is
generative only where \(Q\) rises with it. In GADM-0.2, the organism
contribution to \(V\) should therefore be gated by \(Q\), not just added.

### 3.2 Selection \(\sigma\)

Organisms add a fifth selection channel:

\[
\sigma = w_m M + w_s Sci + w_c Com + w_p Pub + w_o O
\]

where \(O\) is organism-mediated selection: ranking, recommendation,
screening, automated review. Observers and multi-dimensional fitness from the
evolution layer make \(O\) potentially high quality. However, \(O\) is also
exactly the *DiscoveryControl* term in \(\dot P\).

**Hypothesis M3.** The quality of \(O\) and its contribution to \(P\) are
determined by the same variable: who configures the observers and membranes.

### 3.3 Preservation and reuse \(\rho\)

Inheritance of learned structure, including failures, is a preservation
mechanism with its own conversion efficiency into \(X\). Whether it counts as
\(I_X\) (commons) or feeds \(L_X\) (enclosure) depends on lineage licensing.

### 3.4 Claims on output \(I\)

GADM-0.1 decomposes claims as \(I = W + \Pi_p + D_s\). Organisms push
\(W/Y\) down, and route their returns to whoever owns the lineage. That
concentrates \(\Pi_p\) unless ownership is diffuse.

The post-work participation question in INTENT.md suggests a fourth term:

\[
I = W + \Pi_p + D_s + C_p
\]

where \(C_p\) = **participation claims**: claims arising from supplying what
organisms cannot supply themselves, such as objectives, judgment, consent,
evaluation as observers, and governance. Whether \(C_p\) is a real
distribution channel or a relabeled dividend is an open research question
(WS-14).

### 3.5 Domination \(\dot P\)

Two terms to add explicitly:

\[
\dot P = \dots + MembraneControl + LineageConcentration
- MembranePortability - LineageDiffusion
\]

The success → resources → control loop in GADM-0.1 § 13 is sharper for
organisms: a dominant lineage learns from the most use, so it improves fastest
and attracts the most use.

---

## 4. Variables GADM-0.1 lacks

| # | Candidate | Type | Why it is needed |
|---|---|---|---|
| 1 | **Attention headroom \(A\)** | stock | Human attention is the reference scarce resource. It drives adoption \(M\), observer quality in \(O\), participation claims \(C_p\) and cohesion \(S\). Organisms draw on it as premium energy. GADM-0.3 already plans to "model adoption and attention separately". |
| 2 | **Organism capability \(K_{\Omega}\)** | stock | Learning-with-use and lineage dynamics (§ 1). |
| 3 | **Compute \(C\)** | flow or stock | The metabolic link between \(K_{\Omega}\) and \(E\). It makes the energy cost of machine cognition explicit instead of hidden in \(f(E)\). |
| 4 | **Membrane governance distribution** | parameter of \(P\) and \(G\) | Who configures what organisms sense and do. The main control interface in the organism lens. |
| 5 | **Participation claims \(C_p\)** | flow | The post-work participation mechanism (§ 3.4). |
| 6 | **Authority boundary** | constraint, not a stock | Learning is distinct from authority. Like VFAC's constitutional constraints, it bounds optimization and must not be traded off inside it. |

### 4.1 Consequence for VFAC

If \(A\) becomes a stock, it is a candidate factor in VFAC:

\[
VFAC = H^a X^b G^c N^d Q^e S^f E^g (1-P)^h A^i
\]

\(A\) must mean **unclaimed, self-directed attention**, not attention
captured by products. Otherwise maximizing VFAC would reward attention
extraction, the opposite of the intent. This is a concrete Goodhart risk
to record in the metrics catalog.

---

## 5. Proposed scenario dimension

INTENT.md calls for an information-organism dimension in a later scenario set.
Using the variables above, three axes are enough to start:

1. **Lineage ownership:** concentrated ↔ diffuse
2. **Membrane control:** operator-controlled ↔ user/community-controlled and portable
3. **Attention regime:** extraction-funded ↔ protected headroom with participation claims

Crossing these with the six GADM-0.1 scenarios (for example, *Attention
oligopoly* is concentrated / operator-controlled / extraction-funded) gives a
first organism-aware scenario grid.

---

## 6. Open questions

1. Does learning-with-use make \(K_{\Omega}\) a natural monopoly in the
   absence of lineage diffusion, and what interoperability level breaks that
   loop?
2. Is \(C_p\) measurable separately from \(D_s\)?
3. Can organism-mediated selection \(O\) raise \(Q\) without also raising
   \(P\)?
4. What is the right unit of \(A\): time, quality-weighted time, or reachable
   self-directed activities?
5. Where does the organism lens's "viability precedes optimization" map
   formally: onto VFAC's constitutional constraints, or onto the floors of
   individual stocks?