On Metacognition: Concept, Boundary, Hierarchy and Closure Conditions
Metacognition is traditionally defined as “cognition about cognition”. Within psychology, cognitive neuroscience and artificial intelligence, this has long remained a phenomenal description, lacking a complete definition with clear boundaries, stratified functions, interface specifications and termination conditions for self-referential closure. This paper first distinguishes the limitations of historical formulations of metacognition. It then presents a complete working definition of metacognition, separating metacognition from adjacent constructs including ordinary reflection, memory retrieval, confidence calibration, chain-of-thought reasoning and self-supervision. The internal hierarchy of metacognition is unpacked, with input, output and core operations defined for each layer. Closure rules and recursive termination conditions for four-level self-referential closure are elaborated. A distinction is drawn between the metacognitive axiom set and the engineerable behavioural definition of metacognition, alongside their logical relationship: the axiom set generates the definition, and an engineered system satisfies the definition without necessarily carrying the axiom set explicitly. Finally, necessary conditions for metacognitive systems, falsifiability criteria, and two architectural schemas for AGI endowed with metacognition are provided.
Authors
- dongnan li
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-13
- DOI
- https://doi.org/10.5281/zenodo.24185243
- Primary Topic
- Philosophy and Theoretical Science
- Type
- preprint