Metacognitive Arbitration as Uncertainty Compression in Multi-Agent Language Models

Mafizur Rahman, Lijun Qian
Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence, PMLR 337:5623-5642, 2026.

Abstract

Multi-agent reasoning and metacognitive strategies are widely used to improve large language model ({LLM}) performance, yet the functional role of metacognition remains unclear. Prior methods often treat metacognition as a mechanism for generating better or more diverse reasoning. In this work, we argue that metacognition primarily functions as an information routing and compression mechanism rather than a generator of new reasoning content. We introduce MC-MAS, an inference-time framework that separates problem solving from metacognitive arbitration, where independent solvers propose candidate answers and an arbiter critiques and consolidates their outputs. Using routing-centric metrics: semantic novelty, entropy reduction, and overconfident errors across four reasoning benchmarks and multiple model settings, we show that MC-MAS introduces limited novelty but consistently reduces uncertainty and overconfidence, with accuracy gains that are modest and dataset-dependent. These results indicate that reliable improvements arise from structured arbitration rather than reflection or increased sampling alone.

Cite this Paper

BibTeX
@InProceedings{pmlr-v337-rahman26a, title = {Metacognitive Arbitration as Uncertainty Compression in Multi-Agent Language Models}, author = {Rahman, Mafizur and Qian, Lijun}, booktitle = {Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence}, pages = {5623--5642}, year = {2026}, editor = {Perković, Emilija and Malinsky, Daniel}, volume = {337}, series = {Proceedings of Machine Learning Research}, month = {17--21 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v337/main/assets/rahman26a/rahman26a.pdf}, url = {https://proceedings.mlr.press/v337/rahman26a.html}, abstract = {Multi-agent reasoning and metacognitive strategies are widely used to improve large language model ({LLM}) performance, yet the functional role of metacognition remains unclear. Prior methods often treat metacognition as a mechanism for generating better or more diverse reasoning. In this work, we argue that metacognition primarily functions as an information routing and compression mechanism rather than a generator of new reasoning content. We introduce MC-MAS, an inference-time framework that separates problem solving from metacognitive arbitration, where independent solvers propose candidate answers and an arbiter critiques and consolidates their outputs. Using routing-centric metrics: semantic novelty, entropy reduction, and overconfident errors across four reasoning benchmarks and multiple model settings, we show that MC-MAS introduces limited novelty but consistently reduces uncertainty and overconfidence, with accuracy gains that are modest and dataset-dependent. These results indicate that reliable improvements arise from structured arbitration rather than reflection or increased sampling alone.} }
Endnote
%0 Conference Paper %T Metacognitive Arbitration as Uncertainty Compression in Multi-Agent Language Models %A Mafizur Rahman %A Lijun Qian %B Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence %C Proceedings of Machine Learning Research %D 2026 %E Emilija Perković %E Daniel Malinsky %F pmlr-v337-rahman26a %I PMLR %P 5623--5642 %U https://proceedings.mlr.press/v337/rahman26a.html %V 337 %X Multi-agent reasoning and metacognitive strategies are widely used to improve large language model ({LLM}) performance, yet the functional role of metacognition remains unclear. Prior methods often treat metacognition as a mechanism for generating better or more diverse reasoning. In this work, we argue that metacognition primarily functions as an information routing and compression mechanism rather than a generator of new reasoning content. We introduce MC-MAS, an inference-time framework that separates problem solving from metacognitive arbitration, where independent solvers propose candidate answers and an arbiter critiques and consolidates their outputs. Using routing-centric metrics: semantic novelty, entropy reduction, and overconfident errors across four reasoning benchmarks and multiple model settings, we show that MC-MAS introduces limited novelty but consistently reduces uncertainty and overconfidence, with accuracy gains that are modest and dataset-dependent. These results indicate that reliable improvements arise from structured arbitration rather than reflection or increased sampling alone.
APA
Rahman, M. & Qian, L.. (2026). Metacognitive Arbitration as Uncertainty Compression in Multi-Agent Language Models. Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence, in Proceedings of Machine Learning Research 337:5623-5642 Available from https://proceedings.mlr.press/v337/rahman26a.html.

Related Material