08-field-theory-of-value.md

Toward a Field Theory of Value (speculative)

The frontier extension of the thesis: take the discrete multi-agent theory (03, 04, 05) to a continuum, so that value becomes a field over a population of agents, demand-shocks become waves, alignment becomes interference, and the spontaneous coordination of goals becomes a phase transition. The ambition is to study intelligence the way we study physics — as fields, symmetries, and collective phenomena.

Honesty first. This is a research program, not a result. What is rigorous here is the borrowed physics — the equations below are standard (telegrapher's equation, Frank elastic energy, Goldstone's theorem, Toner–Tu active hydrodynamics). What is speculative is the mapping of value/economics onto them. The whole document is a construction motivated by analogy plus one real anchor (active matter); §9 is blunt about what would make it science rather than physics-envy.

Status update. The mapping is no longer pure analogy: 10 derives the telegrapher and Toner–Tu forms by coarse-graining the discrete value dynamics, and sim/field/dynamic/ confirms both signatures emerge from a toy agent value economy (7/7). The first real-agent test (11, an LLM-agent economy) was then attempted and did NOT clear the gate — a pre-registered negative for the transition, inconclusive (confounded) for the wave, obstacle = LLM token bias + a weak noise knob. So the honest status is: derived + emergent in a toy economy, not validated on real agents. Read the inline > Derived… notes in §3, §6 and the §10 roadmap (rungs 2/2b done, rung 3 attempted-not-cleared) for what changed.

0. The one anchor that keeps this honest: active matter

There is already a mature physics of many goal-directed agents that align with their neighbors: active matter — the Vicsek model and its continuum limit, the Toner–Tu equations, which describe flocks, swarms, and self-propelled particles as a hydrodynamic field theory with an orientational order parameter. Active matter already has: an alignment order parameter, spontaneous symmetry breaking (the flocking transition), density waves, dissipation, and dispersion relations. A field theory of value is active matter plus an economic sector (resource + price), plus the is/ought source asymmetry of 05. That lineage is what makes "value is a wave" a physics statement rather than a metaphor — we are not inventing the field theory of aligned agents; we are extending one that works.

1. From agents to fields

Replace the discrete agent index a with a continuous position x in a base space M (a social/network space), evolving in time t. The discrete objects become fields:

Discrete (03–05)	Continuum field over M
goal covector k_a (the frame)	goal-director field n(x,t), a unit vector — the order parameter
resource/budget E_a	resource density e(x,t) (the conserved substrate)
shadow price λ_a=K_a/E_a	price field π(x,t) — the carrier/potential
desire/demand	desire density ρ(x,t) — the source term
value V_a=Σk_i\ln e_i	value density v(x,t) — what is extracted, not conserved

The order parameter is n (where goals point), exactly as in a flock or a nematic. Alignment between two regions is n(x)\cdot n(x'); its sign is the constructive/destructive interference of desires.

2. Three structural inputs, read off the discrete theory

1. Misalignment is gradient energy. In 03, alignment cos θ set positive- vs negative-sum. In the field, that becomes the Frank elastic energy of the director field, F_{\text{align}} = \tfrac{J}{2}\int_M |\nabla n|^2\,dx. Smoothly aligned goals store no energy; gradients (disagreement between neighbors) store energy that can propagate or dissipate. Misalignment is literally curvature in the goal field.
2. Price is the carrier. In 03, resource flows down shadow-price gradients until λ equalizes (a thermodynamic equilibration). In the field, this is a transport law: resource current J_e = -D\,\nabla\pi. Price is the potential whose gradients move the substrate.
3. Value is exergy, so the theory is dissipative. The Second Law (02) says value is created and destroyed, not conserved. A field theory of value therefore cannot be a clean conservative (Hamiltonian) field theory — it is necessarily an open / non-equilibrium / active one, with a dissipation term. This is not a defect; it is the same reason active matter is non-equilibrium physics.

3. The wave equation, and where waves actually come from

Combine resource transport (∂_t e = -\nabla\!\cdot J_e = D\nabla^2\pi) with the market relation that price responds to scarcity and demand, with a response lag τ (agents update beliefs/prices with delay — 05). Expanding the lag to second order yields, for the demand/price disturbance φ, the telegrapher's equation:

$$\boxed{;\tau,\partial_t^2 \varphi ;+; \partial_t \varphi ;=; v^2,\nabla^2\varphi ;+; \text{source}(\rho);,\qquad v=\sqrt{D/\tau}.;}$$

This is exactly the "driven, dissipative wave" of the informal picture, now an equation with a name. Its character depends on a single ratio:

• Overdamped regime (lag small, \tau → 0): reduces to the diffusion equation ∂_t φ = v^2τ\nabla^2φ — demand-shocks spread and damp, no oscillation.
• Underdamped regime (lag large): genuine waves — demand-shocks propagate at speed v=\sqrt{D/τ} (a "speed of value") and oscillate as they damp.

Dispersion relation (\varphi \sim e^{i(q\cdot x-\omega t)}): \tau\omega^2 + i\omega - v^2 q^2 = 0, so $$\omega(q) = \frac{-i \pm \sqrt{4\tau v^2 q^2 - 1}}{2\tau}.$$ There is a crossover wavenumber q^\* = 1/(2v\sqrt{\tau}): short-wavelength (local) demand-shocks propagate as waves; long-wavelength (society-wide) ones diffuse and damp. This is the central falsifiable prediction — a real demand-propagation experiment on a network should show this crossover, or the field picture is wrong.

Now derived, not posited (10 §2). Coarse-graining substrate conservation plus a relaxational resource current (the transmission-line / Maxwell–Cattaneo route) yields this exact equation with τ=L/R, D=1/(RC), v=1/\sqrt{LC} — where C is the storage elasticity \partial e/\partial\pi, R\sim1/\kappa the disequilibrium dissipation, and L the reallocation inertia. The wave exists iff L>0 (finite reallocation response time); with instantaneous reallocation the theory predicts only diffusion. So τ is no longer free — it is a measurable condition.

This is the formal content of "an individual's desire creates a wave in society": a localized source \rho in the telegrapher's equation radiates a damped wave through the price field, at speed \sqrt{D/\tau}.

4. The order-parameter sector: alignment as collective dynamics

The goal-director field obeys active-matter (Toner–Tu-like) dynamics — relaxation toward local alignment plus advection plus noise:

$$\partial_t n + (u!\cdot!\nabla)n = \Gamma\big(\nabla^2 n - (n!\cdot!\nabla^2 n),n\big) ;+; \beta,\Pi_n\nabla\pi ;+; \boldsymbol\eta,$$

where \Gamma relaxes goals toward neighbors (the alignment coupling), \beta\,\Pi_n\nabla\pi re-aims goals toward where value/price is high (agents chase value, projected orthogonal to n to keep |n|=1), and \boldsymbol\eta is the noise of individual idiosyncrasy. This is the field version of "agents adjust their goals under control + selection" (05 §3; 07).

5. The exciting prediction: collective goals are a phase transition

When the alignment coupling \Gamma beats the noise \boldsymbol\eta, the director field spontaneously orders: a population of scattered desires condenses into a coherent collective goal. This is a genuine phase transition — the flocking / order–disorder transition of active matter — and it is the field-theory account of movements, manias, paradigm shifts, fashions, and bubbles: society condensing onto a shared direction of value.

Two consequences carry real predictive content:

• Criticality. Near the transition, correlation lengths diverge and fluctuations become scale-free — predicting power-law demand cascades (virality, fat-tailed adoption) as a critical phenomenon, not an accident.
• Hysteresis / metastability. First-order versions predict that collective goals, once formed, resist dissolving (lock-in) — the value analog of supercooling.

6. The is/ought asymmetry as a mass spectrum (the deepest restatement)

05 §0 found that beliefs have a world-given target and goals do not. In field language this is a statement about the mass spectrum of the two fields:

• The belief field is explicitly pinned to reality q — an external source acts as a symmetry-breaking field, i.e. a mass term. Beliefs are massive: pull them and they spring back to truth.
• The goal field has a continuous symmetry (rotations in goal space) that is spontaneously broken when the population aligns — with no external field pinning the direction. By Goldstone's theorem, the broken symmetry produces gapless (massless) modes: long-wavelength re-orientations of the collective goal that cost vanishing energy.

$$\boxed{;\text{Beliefs are massive (pinned by reality). Goals are Goldstone (massless, free to swing).};}$$

The slow, society-wide value-waves we experience as cultural and ideological drift are the Goldstone modes of the goal field — and they are soft precisely because goals have no world-given target. Hume's is/ought gap becomes: the belief field has a mass, the goal field does not. This is the single most striking thing the field picture buys, and it is a precise, structural claim.

Derived and sharpened (10 §3.3): the control term of 05 §3 coarse-grains to an external field on the director, i.e. a mass m^2\propto\gamma (the control gain). So goals are Goldstone (massless) only when uncontrolled; alignment design is literally the addition of a mass term. Confirmed emergently — control rounds the flocking transition (§5), the simulated signature of a mass — in sim/field/dynamic/ (C3).

7. What is conserved, and what is not

The truly conserved object is the substrate (resource/free energy): a continuity equation ∂_t e + \nabla\!\cdot J_e = 0 in a closed system. Value is the part extracted along goal-gradients — a source/sink term, created where n aligns with resource flow and dissipated where it does not. The Lagrangian (action) formulation works only for the conservative sector; the dissipation (the Second Law) is non-Lagrangian — it needs a Rayleigh dissipation function or an open-system (non-Hermitian) treatment. Restating: a field theory of value is irreducibly a theory of active, driven matter — there is no Hamiltonian for it, and that is forced by the Second Law of Value, not a modeling shortcut.

8. "Intelligence as physics" — what the program actually claims

Stated carefully, the claim is not that minds are literally fields, but that populations of goal-directed agents admit a statistical field theory the same way populations of spins, molecules, or birds do. Its observables are collective: order parameters (alignment), waves (demand propagation with a dispersion relation), phase transitions (collective-goal formation), critical exponents (cascade statistics), and a characteristic asymmetry (massive beliefs, massless goals). Intelligence-at-scale would then have a phase diagram — regimes of coherent collective purpose vs. incoherent individual drift, with measurable transitions between them. That is what it would mean to study intelligence as we study physics: not metaphor, but order parameters and dispersion relations you can measure.

9. Where this breaks — the honest guard against physics-envy

The history of "physics of society" is littered with equations mapped onto social systems that did not obey them. This program is worth nothing until it earns the analogy:

• Locality and symmetry are assumed, not established. Field theory needs local interactions and a real symmetry group. Social/agent networks are long-range, heterogeneous, and may have no clean symmetry — in which case the field reduction is invalid and only the discrete theory survives.
• The mappings are constructions, not derivations derivations conditional on explicit assumptions. (Updated, 10.) The telegrapher and Toner–Tu forms now derive by coarse-graining the discrete value dynamics, with the coupling constants expressed as functions of agent-level quantities (τ,D,v from storage elasticity / dissipation / reallocation inertia; Γ,J,β from imitation rate, density, range, replicator sensitivity). What remains assumed is no longer the equations but two falsifiable conditions: reallocation inertia L>0 (else the wave is pure diffusion) and local, rotationally- symmetric, motile goal imitation (else the gradient expansion / spontaneous ordering fails). The honest gap moved from "are these the right equations?" to "do real agent networks meet these conditions?".
• The decisive test is a measured dispersion relation or a measured transition. The program becomes science the day someone measures, in a real agent population or market, either the §3 crossover q^\* (waves at short scale, diffusion at long scale) or the §5 flocking transition with its critical exponents. Absent that, this is a suggestive structure, explicitly labelled as such.
• No validated empirical content on real agents. A first real-agent test was attempted (11, dynamic LLM-agent economy) and did not clear the gate: a pre-registered negative for the transition and an inconclusive-confounded result for the wave, with the obstacle identified as LLM token bias + a weak noise knob (not, yet, the physics). So this remains the most speculative document in the repository: derived (10) and emergent in a toy economy (sim/field/dynamic/), but not validated on real agents. Read it as the research frontier, not a claim.

10. Roadmap

This is the long-horizon extension, gated behind the nearer empirical work (06 scale-up). The tractable first steps, in order:

1. ✅ Simulate it — DONE (sim/field/, 5/5). The lattice limit does have the claimed phenomenology: a demand-shock spreads ballistically (σ²∝t², a wave) at large lag and diffusively (σ²∝t) at small lag, with wave speed obeying the telegrapher law v∝1/√τ to 1%; and a population of goal-directors undergoes the Vicsek order→disorder phase transition (m: 0.99→0.03). This clears the first gate — does the continuum even have the phenomenology? — yes. It is an internal-consistency check, not evidence about real agents (next steps remain). 1b. ✅ First real-data touch — DONE (sim/field/v2_geometry.py). The field theory's order-parameter assumption holds on a real 10-agent population: across 3 domains, strong positive cross-agent alignment (correlation 0.32–0.51) and a dominant shared-competence axis (λ₁/Σλ 0.41–0.51), tens of σ above the independence null (z = 32–51). Real agents have the low-dimensional alignment geometry the field assumes — and this static geometry explains the v1/v2 R5 result (the population is deep in the constructive-interference/aligned regime, so there is little anti-correlated diversity for pooling/pricing to harvest). Scope: supports the foundations; NOT a test of the wave (§3) or the transition (§5) — those need evolving goals, which static classification lacks.

2. ✅ Pin the couplings — DONE (10). Coarse-graining the discrete value dynamics (04–05) derives both forms: the telegrapher equation from substrate conservation + a relaxational resource current (τ=L/R, D=1/RC, v=1/\sqrt{LC}), and Toner–Tu from local replicator-weighted goal imitation (the Bertin–Droz–Grégoire route; Γ\propto J_{\rm align}\rho r^2), with control = a mass m^2\propto\gamma grounding §6. The constants are now functions of agent quantities, not free. Conditions replace posits: the wave needs reallocation inertia L>0; the transition needs local, symmetric, motile imitation. Either could fail on real networks — that is now the explicit, falsifiable gap. 2b. ✅ Emergent confirmation in a dynamic toy economy — DONE (sim/field/dynamic/, 7/7). Toy agents with only the micro-rules (resource shipped toward value with a flow lag; goals imitating resource-capturing neighbours) — no field equation integrated — reproduce the demand wave with the predicted dispersion crossover q^\* (4/4) and the collective-goal order→disorder transition with a susceptibility peak and the control-as-mass rounding (3/3). The wave needs reallocation inertia and the transition needs motility (Mermin–Wagner kills it on a fixed lattice) — exactly the derived conditions. This is the bridge between the pure-physics lattice (rung 1) and real agents; still not real-agent data (rung 3).

3. ⏳ Real-agent test — ATTEMPTED, gate NOT cleared (11; sim/field/real/). The decisive test was run as a dynamic LLM-agent economy (qwen2.5 0.5b/1.5b choosing each agent's niche each round, pre-registered thresholds). Outcome: a pre-registered NEGATIVE for the collective-goal transition (no order→disorder collapse within the frozen noise range on either model; exploratory high-temp shows at most a soft crossover, no disordered phase) and an INCONCLUSIVE-confounded result for the wave (the shock fell on the model's most-favoured niche-token). The binding obstacle is LLM token bias (small models emit favourite niche-tokens regardless of value, confounding both the order parameter and the value-response) plus a weak noise knob (sampling temperature is not faithful Vicsek alignment-noise — the neighbour majority stays visible in-prompt). Neither a validated positive nor a clean refutation: the obstacle is the LLM-as-agent, not (yet) the physics. 3b. ⚠️ Bias-controlled re-test (Rung 8) — CLEAN NEGATIVE for spontaneous order (11 §7; sim/field/real/PREREGISTRATION_rung8.md). Per-agent symbol-randomisation removes the token-bias artifact (neutral-context order falls 0.40–0.51 → 0.06) while the LLM still makes the real value-decision (noise = its own temperature; no externally-imposed alignment-noise — guardrail). With the artifact controlled the ordered phase vanishes (m≈0.18–0.39 at all temperatures; m(0.2)=0.18≈random; 0/4). So Rung-7's "collective goal" was the shared token bias acting as a Schelling focal point, not spontaneous flocking: an external control field γ does induce order (m(0.2): 0.18→0.51), but agents do not spontaneously break symmetry. The §5 spontaneous transition is not supported on real agents (the §6 field-induced order is, weakly). The wave: reward-following genuinely works once bias-controlled (locus adopts ≈0.7) but the shock does not propagate (no spontaneous coordination to carry it). Per the CAP we did not escalate past 1.5b or supply a focal point. Keeper: small-LLM value-coordination is focal-point-driven; the apparent order parameter is, uncontrolled, substantially a token-bias artifact. (Gate still open; doc 08 §5 not supported on real agents at this scale.)

4. Only then: the analytic field theory (renormalization, the phase diagram, the critical exponents).

The prize, if it survives: a phase diagram of collective intelligence — and the field-theoretic statement that the soft modes of shared purpose exist because goals, unlike beliefs, answer to nothing outside themselves.