h t t p s : / / f i t e l s o n . o r g / p i k s i / p i k s i _ 2 2 / a g m . p d f

Ingest: On the Logic of Theory Change

Type: kb/sources/types/ingest-report.md

Classification

This is a formal scientific paper: it defines contraction and revision functions, proves their properties, and establishes representation theorems rather than reporting an empirical study. Author: Carlos E. Alchourrón, Peter Gärdenfors, and David Makinson are the originators of the AGM framework, and the paper appeared in The Journal of Symbolic Logic.

Summary

The paper asks how a logically closed theory should change when a proposition is removed or when an inconsistent proposition is added. It defines partial meet contraction as the intersection of a selected nonempty family of maximal subsets that do not imply the proposition being removed, then defines revision from contraction through the Levi identity. Its central result is a representation theorem: for theories, the basic Gärdenfors contraction postulates characterize partial meet contraction functions, with corresponding revision postulates; additional restrictions on selection yield supplementary postulates. This makes the paper a precise account of admissible change to an existing theory under a fixed consequence operation, not an account of how to discover theories or establish that revised beliefs are empirically reliable.

Quotes

No source quotes have been retained yet.

Connections Found

The paper is a technical basis for separating formal theory change from epistemic warrant. It supports A claim's warrant does not determine its fit in a working theory: the postulates constrain how membership in a theory changes, while leaving the independent warrant of retained or introduced propositions unsettled. It also compares with Machinery persists by warrant, not position, in a reflective loop: AGM supplies formal contraction and revision machinery, whereas that note requires governance by current warrant and empirical consequences beyond formal coherence.

Learning Claims (our opinion)

On the paper's terms, adaptation is an operation over a theory already closed under a fixed consequence relation. Contraction removes a proposition while preserving selected maximal portions of the prior theory; revision first contracts enough to accommodate the new proposition and then adds it. Relative to Theory refinement, this supplies a rigorous family of change operators and preservation constraints, but only a partial mapping. AGM's selection function expresses which remainder sets are preferred without explaining how evidence diagnoses a fault, generates candidate repairs, or selects among them. Its representation theorems establish equivalence between postulates and formal constructions, not that either construction learns reliable theories from empirical cases.

Extractable Value

  1. Use the postulates as a precise citation for contraction and revision. Contraction removes a proposition from an existing theory while preserving closure and specified rationality properties; revision incorporates a proposition consistently through contraction plus expansion. [just-a-reference]
  2. Keep formal admissibility separate from warrant. The representation theorem says which operations satisfy the postulates, but it does not show that retained beliefs are true, well supported, or useful. This sharpens the distinction in A claim's warrant does not determine its fit in a working theory. [quick-win]
  3. Treat the selection function as an unresolved governance input. Partial meet contraction depends on selecting the most important maximal subsets, but the paper's construction does not supply an empirical or discovery procedure for making that choice. [deep-dive]
  4. Distinguish theory revision from theory discovery. AGM begins with a theory, a consequence operation, and a proposition to remove or add; it does not construct new explanatory content or identify which proposition an observation should overturn. [quick-win]
  5. Use recovery and preservation as comparison points, not universal requirements. These postulates make specific commitments about retaining prior content and treating logically equivalent inputs alike; applying them to natural-language or mixed-form theories requires an explicit interpretation of consequence and equivalence. [deep-dive]

Limitations (our opinion)

The paper is formal rather than empirical. Its conclusions depend on an abstract consequence operation with stated logical properties and on a selection function whose substantive basis is left outside the representation theorem. It does not test whether agents can identify faulty commitments, choose useful remainder sets, interpret natural-language consequences consistently, or improve prediction and action after revision. The framework therefore cannot by itself support claims about discovery, empirical reliability, or successful theory refinement. Its closed-theory setting also does not transfer directly to Commonplace's mixed-form, partly interpreted theories without specifying how consequences and editable parts are determined.

Add this ingest as the formal citation in A claim's warrant does not determine its fit in a working theory, with one sentence stating that AGM characterizes contraction and revision but does not supply discovery or empirical warrant.