Tree F: wetware comparison. Plan
Status 2026-09-30: parked indefinitely. Tree A's ledger failed
calibration, so there is no cross-substrate residual to compare. A
partner could still run the barrier assay on their platform. No line
depends on it. Context: ../../research-plan.md section 2
(Tree F).
Goal
Run the same barrier assay on cultured neurons and on a software model matched for task performance, then compare their residuals under the same ledger. Levin's synthetic-biology origin and the AI programme meet on one bench here and nowhere else.
Hypotheses
- F1 (Seth's biological naturalism, made quantitative): neurons show a larger unpaid residual than silicon on matched tasks.
- F2 (substrate independence): the residual is the same once training signal is matched.
Prerequisites
- An accounting rule. Tree A's ledger v0.1 failed calibration at gate A2 (2026-09-30), so this waits on spine/01 measure M5.
- A barrier harness that accepts an external system via a network interface (Tree B).
- A partner. The candidates are Cortical Labs (Kagan), FinalSpark's Neuroplatform (cloud access to organoids, and the cheapest entry), or an academic lab with DishBrain-style setups.
- An ethics review. Whether organoids and neuron cultures can be sentient is a live debate (Bayne, Birch, 2026), and the group's rule against welfare claims applies here twice over.
Steps
- Define the shared task. The DishBrain Pong or Doom paradigms already have the shape of a barrier task. Inputs, outputs, perturbations and scoring are specified identically for wet and dry.
- Partner outreach (todo-operator.md).
- Software side. Train a small model to the same task performance under the same feedback protocol, and run the ledger.
- Wet side. The partner runs the protocol and we receive traces.
- Compare residuals, following a pre-registered analysis.
Decision gates
- If there is no partner within six months of starting outreach, the tree stays parked.
- If residuals differ beyond the null, a second wet platform replicates the result before any claim.
Cost
The highest of all the trees. It needs partner time, platform fees (FinalSpark charges per access slot) and an ethics review. Not before year two.