TheoryTHY-2026-0103v0.1
Cross-level triangulation and measurement grades
Neuroscience has no '1 thought = N spikes' conversion; what it has is a cross-level proxy method: behavior → latent cognitive model → neural code → cellular events → physical implementation, each layer with its own units. IPM borrows five principles (level separation, latent inference, population over atomism, encoding–decoding duality, causal perturbation), extends Marr's three levels to five (task, semantic, algorithmic, physical events, thermodynamic), and defines cross-level triangulation for AI as E = (output, semantic, internal trace, ablation, hardware) evidence with a graded confidence in μI from D (behavioral) to A+ (physical-semantic alignment).
Definitions
definitions- Five layers L4 task achievement, L3 semantic/cognitive operation, L2 algorithmic realization, L1 physical events, L0 thermodynamic realization; L4 ≠ L3 ≠ L2 ≠ L1 ≠ L0.
- Cross-level triangulation X_L = (E_B behavioral, E_C cognitive-model, E_N neural, E_P perturbational); for AI X_AI = (E_O, E_S, E_I, E_A, E_H).
- Measurement grades D behavioral / C structured semantic / B internal correlation / A causal internal / A+ physical-semantic alignment.
- Eight borrowing principles: level separation, proxy discipline, model-mediated inference, distributed realization, causal perturbation, scale declaration, trial separation, grounding downward.
Assumptions
assumptions- Latent quantities are scientific when they make observable predictions, have competitors, are falsifiable and accept intervention (the diffusion-decision-model template).
Claims
claims- CognitiveUnit ≠ NeuralEvent ≠ InformationBit ≠ PhysicalOperation; bits/spike ≠ cognitive bits; Decodable ≠ CausallyUsed.
- OutputRate ≠ InternalComputationRate (the ~10 bits/s behavioral throughput is not the brain's computation rate; 1000 output tokens are not 1000 intelligent events).
- EnsemblePerformance ≠ SingleEpisodePerformance; Proxy ≠ Ontology; BiologicalNeuron ≠ ANNNeuron.
Formalisation
formalization- Conf(μI) = F(E_O, E_S, E_I, E_A, E_H) ∈ [0, 1]; report N̂_μ ± uncertainty with Grade_μ, never a falsely precise count.
- ValueOfComputation = ExpectedImprovement − Cost (resource-rational template for μI → Cost → Value).
Predictions
predictions- Confidence that μI^obs ≈ μI^int rises only when behavioral, semantic, internal, causal and physical evidence converge.
Falsification / failure conditions
falsification_conditions- A μI account that survives behavioral evidence but is contradicted by ablation or hardware evidence must lose its grade, not be kept by verbal interpretation.
Known limitations
known_limitations- Methodological borrowing only; no biological unit is equated with an AI unit and no measurement was performed.
Relations
| Source | Relation | Target | Status | ID |
|---|---|---|---|---|
THY-2026-0103 Cross-level triangulation and measurement grades | extends | THY-2026-0102 μI — the minimum intelligent semantic execution unit (candidate theory) | ACTIVE | REL-2026-0369 |
RES-2026-0101 Execution and physical line — what one answer costs in turns, semantic work, energy and computational spacetime | develops | THY-2026-0103 Cross-level triangulation and measurement grades | ACTIVE | REL-2026-0360 |
THY-2026-0104 Energy accounting hierarchy and thermodynamic type safety | extends | THY-2026-0103 Cross-level triangulation and measurement grades | ACTIVE | REL-2026-0370 |
PAP-2026-0103 Paper 03 — From cognition to neurons: how human intelligence is measured across levels | formalizes | THY-2026-0103 Cross-level triangulation and measurement grades | ACTIVE | REL-2026-0400 |
History and provenance
- Canonical URL
- https://evemisslab.com/ai/theory/THY-2026-0103/
- Machine-readable
/ai/theory/THY-2026-0103/index.json- Snapshot
AI-SNAPSHOT-v0.1-fe85b9694a45- Provenance
source- EveMissLab research collection: Intelligence Physical Metrology (真本體論13)
extracted_by- Splice (Claude Code), reading the canonical UTF-8 sources and each package's own reports
extracted_at- 2026-09-11
generator- tools/extract_all.py
claim_boundary- status, evidence level and result type follow the source artifact's own stated claim boundary; nothing is upgraded beyond what the report supports