Experiments

V20: Protocell Agency

V20: Protocell Agency

Period: 2026-02-18. Substrate: Discrete grid world with evolved GRU agents (~3400 params each).

Protocell Grid World16×16 portion — resource patches, agents, observation windows★Resource patchFocal agentSame typeOther type

The Necessity Chain: Membrane → free energy gradient → world model → self-model → affect geometry. Each step necessary, not contingent.

The Necessity Chaineach step necessary, not contingent (V20)1MembraneBoundary separating self from non-selfdiscrete grid bodynecessitates2Free Energy GradientViability requires energy harvestingresource depletion → deathnecessitates3World ModelPrediction of own future energyC_wm = 0.10–0.15necessitates4Self-ModelAgent encodes own state more than environmentSM_sal > 1.0 in 2/3 seedsnecessitates5Affect GeometrySimilarity structure on internal statesRSA ρ > 0.21, eff. rank 5–11
Protocell Agent Architecture (V20–V35)extractinputW₁·tanh(W₂·h)softmax(W_a·h)action→observation loop (ρ wall)MSE(ŷ, y)Grid World32×32, N agents, R resourcesObservationlocal window + compassGRU CoreH=16 or H=32Hidden State heff. rank 5–11Prediction Headlinear or 2-layer MLPAction Head4 dirs + eat + (comm)SGD Stepwithin-lifetime learning

Why leave Lenia: V13-V18 showed the wall is about agency, not signal routing. GRU agents on a discrete grid have bounded local sensory fields (5x5), genuine motor actions (move, consume, emit), and observations shaped by their own actions.

MetricSeed 42Seed 123Seed 7
ρsync\rho_{\text{sync}} (max)0.2300.2320.212
Cwm\mathcal{C}_{\text{wm}} (final)0.1220.1260.152
SMsal\text{SM}_{\text{sal}} (final)1.220.941.50
RSA (final)-0.017-0.0050.031

Wall broken (3/3 seeds). ρsync=0.21\rho_{\text{sync}} = 0.21 from cycle 0. The wall is architectural — action-observation loops are present from initialization. 70x Lenia's 0.003. Self-model emergent in 2/3 seeds (SMsal>1.0\text{SM}_{\text{sal}} > 1.0).

Source code