Tentative theory

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

A tentative theory is a theory put forward as a solution to a problem, which stays open to criticism and revision however well it has survived. The term is Popper's: TT in his schema P1 → TT → EE → P2 stands between a problem and the attempted error elimination that leads to new problems (Popper 1966). The KB uses the term as he does and adds nothing to it.

The KB needs the term because every theory it retains has this status, and the status must be nameable without implying anything about how the theory is stored or edited. Structural properties have their own term, addressable theory.

Scope

  • A status, not a stage. A theory stays tentative even when we no longer feel able to doubt it (Conjectures and Refutations, Chapters 1 and 15). Corroboration can strengthen the grounds for later reliance without making the theory less tentative in this technical sense. Neither corroboration nor changing the theory's representational form ends the status.
  • Independent of the holder. The status belongs to every theory, whatever a system does with it. A theory encoded in fixed instructions remains tentative even if the system is not permitted to revise those instructions. Restrictions on revision and the presence of criticism are separate facts about the system; neither determines tentative status.
  • No required form. The term does not require symbolic formalization, storage outside the model weights, separately editable parts, or a particular consumer. A theory that can only be replaced whole is still tentative. Popper applies the schema to expectations and behaviour as well as to formulated theories; on that wide reading the dispositions held in a model's weights count as tentative theories too. Applying his account of dispositions to weights is our interpretation, not a finding about any particular model, and it does not establish that nothing in a model is formulated. A theory builder works only with the stated ones (its condition 1).
  • Criticism is broader than empirical test. A tentative theory may be criticized by argument, by comparison with a rival, or by testing a prediction. A failed prediction counts against it; surviving a serious empirical test can add to its support.
  • Licenses no particular use. The term does not say how much support permits retention, experimental use, routine reliance, or codification. Those are policy questions a builder answers for each use, collected in a claim without external assessment carries three obligations.

Exclusions

  • Provisional or hesitant in the everyday sense. A theory is not tentative because its author is unsure of it or because approval is pending.

Misuse Cases

  • Treating the term as evidence of quality. A false theory and a true one are tentative alike.
  • Importing structure into the term: requiring parts, storage outside model weights, or a formal language before calling a theory tentative.

Relevant Notes: