Sublinguistic Intelligence
A third epistemic category and its computational encoding. Formally defined, mathematically grounded, empirically validated.
One• The Constraint
The Labeller's Ceiling
Contemporary AI systems share a structural constraint: their performance cannot exceed that of the humans who generated their training signal. This reflects a fundamental partition in the nature of human knowledge itself.
No system trained from articulated or demonstrated knowledge can exceed the outcome quality of the best labeller, regardless of model capacity, training compute, or architectural choice.
Explicit Knowledge
Propositionally representable and communicable through language without decision-relevant loss.
ArticulableYes
DemonstrableYes
TrainingLabels & rules
CeilingBest labeller
Tacit Knowledge
Resists complete articulation but admits transmission through embodied demonstration.
ArticulablePartial
DemonstrableYes
TrainingDemonstrations
CeilingBest demonstrator
Sublinguistic Intelligence
Neither articulable nor demonstrable. Recoverable exclusively through verified long-run outcomes.
ArticulableNo
DemonstrableNo
TrainingVerified outcomes
CeilingBayes-optimal
In any domain where the best practitioners possess Category III knowledge, the labeller's ceiling lies strictly below expert-level performance. No investment in scale can close this gap within existing paradigms.