The Thesis
What we claim, what we've shown, and what follows
The Claim
High-dimensional dynamics, finite interfaces, and physical realization must be analysed together. A useful description states which distinctions it preserves, which interventions it supports, and what its substrate contributes.
The program tests this approach across neural, cellular, computational, and social models. It is a research strategy, not a definition of intelligence or a universal dimension threshold.
Why external observers can't keep up
An observer only receives the distinctions carried by its measurement channel. A binary test, for example, cannot preserve an arbitrary continuous source state without additional structure, repeated measurements, or a task-specific sufficient statistic.
Generic reconstruction can become exponentially expensive with dimension, but structure matters: sparsity, symmetry, dynamics, and the target question can reduce the burden. Every impossibility or lower bound therefore needs a stated source class, observation map, noise model, and success criterion.
Observation reallocates distinguishability
A measurement or representation selects which distinctions remain accessible. Whether states are identified, distorted, or preserved depends on the map, its noise model, its range, and the question being asked.
Dimensional reduction alone does not force continuous dynamics to become discrete. A monotone-radius planar spiral can preserve every phase–epoch state. When its radial range is bounded, however, successive history records crowd together near the boundary and become harder to distinguish through fixed readout noise. That is a trade-off in the chosen record map, not a universal dimensional threshold.
Biology operates in the gap
Biology supplies demanding test cases for this framework:
Thermodynamics
Landauer's bound applies to logically irreversible operations such as memory reset, not to every bit or clock cycle. Biological and digital systems can both defer or avoid erasure; their efficiency has to be compared using actual devices, tasks, and error budgets.
Timing Inaccessibility →Timing
Synchronization can impose a timing bottleneck in the paper's modular-oscillator model. Under its selected neural parameters and registration rule, the model reproduces a 30–50 ms order of magnitude. That fit does not establish that phase alignment causes visual binding.
Coherence Time →Development
The same genome can support different trajectories because development is path-dependent and environmentally coupled. State-space language can organize that history, but it does not replace a causal developmental model.
Intelligence as High-Dimensional Coherence →Coupling creates joint observables
Coupling can create collective modes and stable cross-dependencies that are absent from the dynamics of either isolated component.
Correlation is mathematically defined for independent variables too; independence constrains it rather than making it undefined. After coupling, a nonzero phase relation or collective mode may become estimable and causally useful. The model must specify the joint observable and show that it improves prediction or intervention.
How to think about emergence
An emergent description earns its place when a joint variable captures stable behaviour that isolated variables do not capture economically. The distinction is operational:
- Before coupling: the joint distribution factorizes under the model, and no stable phase relation is induced.
- After coupling: cross-dependencies and collective modes may become estimable and intervention-relevant.
The interaction changes the joint dynamics. “Emergence” names the useful higher-level variable; it does not excuse us from deriving or measuring the coupling that supports it.
Implications
For AI
An algorithm fixes an equivalence class at a chosen causal grain. That description is not yet a realization: phase, material memory, noise, contact, or environmental coupling may still change intervention responses.
Whether a digital controller omits a task-relevant physical distinction is therefore an empirical, mapping-specific question—not something decided by a universal dimension threshold.
For Consciousness
Physical realization may matter to consciousness, but dimensionality is not a consciousness test. Noise, synchrony, and field structure become relevant only when changing them changes the system's causal organization or available interventions.
Stochastic resonance is a useful example: noise can participate constitutively in a physical operation rather than merely corrupt an abstract symbol. That still does not, by itself, establish irreversibility or experience.
For Science
Tests are always made through specified instruments, variables, and decision rules. A null result can leave models underdetermined when the channel discards distinctions on which they differ.
The response is to state the measurement map, identify its equivalence classes, and design new interventions where rival models predict different outcomes. Limits must be demonstrated for a stated protocol rather than inferred from dimensionality alone.
Limits of Falsifiability →Start Here
Suggested reading order, depending on your background:
For the math/physics reader
- Intelligence as High-Dimensional Coherence — the program hypothesis
- Limits of Falsifiability — what observers can't reach
- Timing Inaccessibility — transient coupling versus durable records
For the biology/neuro reader
- Intelligence as High-Dimensional Coherence — the thesis
- Timing Inaccessibility — transient coupling versus durable registration
- Coherence Time — timing constraints on neural coordination
For the philosophy reader
- Limits of Falsifiability — epistemology of measurement
- Intelligence as High-Dimensional Coherence — graded control across scales
- Quantum Mechanics Without the Math — accessible intro
For the AI/ML reader
- Intelligence as High-Dimensional Coherence — the program's biological hypothesis
- Limits of Falsifiability — measurement maps and underdetermination
- The Meta-Time Spiral — phase, history, and noisy records
See the papers page for the full list organized by role, or explore the interactive simulations.