UTE needs uncertainty with semantics: not only “a score”, but a model of what the score means (priors, likelihood, calibration),
how uncertainty propagates through inference, and how it affects planning and revision.
What exists today
Today, AGISystem2 can report confidence-like metrics and can run hybrid (HDC + symbolic validation) pipelines, but it does not have
a probabilistic semantics layer (no priors/likelihood, no posterior computation, no consistent semantics for uncertainty).
UTE requirement: probability as a first-class citizen
Priors
Represent prior beliefs about facts/mechanisms and the reliability of sources.
Likelihood
Define how measurements and observations support (or refute) hypotheses.
Inference
Compute posteriors (exact or approximate) and expose a provenance-friendly explanation.
Proposed extensions (research direction)
- Probabilistic assertion types in the DSL/runtime (e.g. distributions, bounds, confidence intervals).
- Probabilistic inference module integrated with the proof/evidence system (proof-like “inference traces”).
- Calibration tests as evaluation suites (do probabilities mean what we claim they mean?).
Spec
This page is summarized and formalized in DS36.