Metacognition
Metacognition is the capacity to monitor, evaluate, or regulate one's own cognitive processes.
Introduction to Generative AI, Second Edition
Metacognition is the capacity to recognize, monitor, or think about one's own mental processes.
- How it is used
- Frames consciousness and AI sentience through the question of whether a system can recognize its own mental processes.
- How deep it goes
- Defines metacognition through conscious awareness but stops short of providing a reliable measurement method.
- Examples
- Uses work-email planning, romantic uncertainty, breathing, knee-tap reflexes, and LaMDA’s self-description.
- What you can do after
- Evaluate claims about AI consciousness while distinguishing fluent responses from demonstrated self-awareness.
Read how Introduction to Generative AI, Second Edition explains metacognition →
AI Agents in Action, Second Edition
Metacognitive monitoring is the ongoing assessment of an agent's own knowledge and confidence while producing answers or taking actions.
- How it is used
- Monitors an agent’s confidence, progress, and knowledge limits so it can avoid confidently unsupported answers.
- How deep it goes
- Defines three monitoring functions and shows a boundary-scoring implementation; does not establish that verbal confidence alone is reliable.
- Examples
- Uses confidently wrong production answers and a workspace-based boundary check using retrieval relevance, memory hits, and confidence.
- What you can do after
- Build it from scratch as workspace-based monitoring that checks confidence, detects stalled progress, and recognizes knowledge boundaries.
Read how AI Agents in Action, Second Edition explains metacognition →
Related concepts
- model calibrationhas part · Model calibration is the assessment or adjustment of a model’s outputs, scores, or confidence so they correspond more reliably to observed outcomes or a reference standard.
- knowledge boundary detectionhas part · The identification of whether an agent has sufficient information to answer a question reliably.
- stagnationhas part · Stagnation is a search condition in which successive iterations or generations produce little or no meaningful improvement toward a better solution.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.