Adaptive Metacognition in AI: From Monitoring to Learned Control
Proceedings of the AAAI Symposium Series · 06.11.2026 · 10 (1) · 105-109
DOI: 10.1609/aaaiss.v10i1.43396 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Symposium Series, online erschienen am 06.11.2026. Verfasst von Brendan Conway-Smith.
Abstract (Originalsprache)
Current AI systems possess limited metacognitive capabilities such as confidence estimation, error detection, and self-critique. These capabilities often remain fragmented rather than integrated into adaptive control. As a result, systems may persist with ineffective strategies or waste resources on additional reasoning that is unlikely to help. To bridge this gap, this paper proposes an architecture that links internal monitoring with learned control. Drawing on human metacognition and computational cognitive architectures, the architecture integrates cues related to fluency, conflict, and progress into signals that guide decisions about whether to persist, switch strategies, verify an answer, seek information, or stop. Outcome feedback allows the system to learn which cue-control mappings improve performance in different contexts relative to their costs. With practice, these mappings could support faster, less demanding control. This framework connects internal processing cues, metacognitive monitoring, control decisions, and learning to provide a more integrated account of adaptive metacognition in AI.
Zitieren
APA
Conway-Smith, B. (2026). Adaptive Metacognition in AI: From Monitoring to Learned Control. Proceedings of the AAAI Symposium Series, 10(1), 105-109. https://doi.org/10.1609/aaaiss.v10i1.43396
Vancouver
Conway-Smith B. Adaptive Metacognition in AI: From Monitoring to Learned Control. Proceedings of the AAAI Symposium Series. 2026;10(1):105-109. doi:10.1609/aaaiss.v10i1.43396
BibTeX
@article{ConwaySmith2026,
author = {Conway-Smith, Brendan},
title = {Adaptive Metacognition in AI: From Monitoring to Learned Control},
journal = {Proceedings of the AAAI Symposium Series},
year = {2026},
volume = {10},
number = {1},
pages = {105--109},
doi = {10.1609/aaaiss.v10i1.43396}
}
EUVIMED