Autonomous theory builder

Type: kb/types/definition.md · Tags: foundations, self-improving-systems, learning-theory

An autonomous theory builder is a theory builder whose theory-building pathway is computationally closed: within the declared boundary and over the assessed horizon, computation supplies every internal theory-building role. Those roles are the ones the boundary rule names: constructing a first theory, interpreting what a theory implies, choosing which part to blame or revise, producing a candidate revision, evaluating a candidate theory, selecting the theory to retain, and extending and repairing the machinery. The KB needs the term because autonomy has to be measurable without pretending that an autonomous learner lives in an environment without people.

Users remain outside the builder when they supply questions, cases, evidence, requirements, observations, or acceptance judgments. Whether a person's judgment is an external acceptance judgment or an internal evaluation is decided by the role rule, not by who makes it: evaluating a candidate theory or selecting its successor is internal, and judging the resulting product against the external task contract is not. An assessment reports each consequential role as human, computational, or joint, and declares the seed and its construction separately from interventions during the assessed run.

Scope

  • Independent of reflection and of extension. See reflective theory builder and extension.
  • Not reliability. Computational performance of a role does not establish its reliability. Warranted autonomy, in the sense of warranted autonomy is bounded by oracle domain, is the condition that the builder's evaluators cover the revisions it accepts with the required confidence; it is a separate assessment.
  • Not objective governance. Autonomy does not determine who may change the objective or what would warrant that change. The declared evidence interface can include externally supplied acceptance judgments while the builder remains autonomous; what licenses an objective change is the second obligation in a claim without external assessment carries three obligations.

Exclusions

  • A builder that depends on a person for diagnosis, admission, or successor selection of its theories, however small the person's share.
  • A computational procedure that is not a theory builder, such as a refinement algorithm run once over supplied cases; autonomy qualifies a builder, it does not create one.
  • Reduction of human presence as such. The target the term serves is warranted autonomy, not fewer people.

Misuse Cases

  • Calling a builder autonomous because users supply only tasks, when a person still evaluates or selects its theories.
  • Calling a builder non-autonomous because people judge its products; that is the external acceptance role.
  • Inferring from autonomy that the objective cannot change, or that its change cannot be licensed.
  • Reading a proof obligation on rewrites as reliability beyond what the proof covers.

Boundary cases

  • FORTE performs its refinement computationally but is not persistent, so it is not an autonomous theory builder.
  • Commonplace today is not autonomous: the operator performs diagnosis, admission, and successor selection for library changes, and these are internal roles under the role rule.
  • The Gödel machine performs every role it has computationally, with warrant within its proof surface; whether it is a theory builder is assessed separately and stays open.

Relevant Notes: