The Neural Basis and Computational Models of Metacognition

Abstract: The ability to reflect on ones own thinking is what makes human cognition "meta." Metacognition, the capability to assess, reflect on, and control first-order cognitive processes, is essential for flexible and adaptive behaviors across various contexts. This review explores the neural mechanisms and computational models underpinning metacognition. The involvement of brain regions, including the insula, precuneus, medial prefrontal cortex, and dorsolateral prefrontal cortex in metacognitive judgments is examined. How distinct regions support both domain-general and domain-specific metacognitive processes is also explored. Furthermore, the neural correlates of metacognitive executive functions, such as error monitoring and cognitive control, are investigated, with a focus on the prefrontal and anterior cingulate cortex and their roles in regulating working memory and performance monitoring. This review also discusses the Bayesian models of human metacognitive processes proposed by Fleming and Daw. Studies on human metacognition have significant implications for the development of artificial intelligence, evidenced by the H-CogAff architecture, revealing how integrating metacognitive frameworks could enhance AIs transparency, reasoning, adaptability, and perception. The findings suggest that investigating the neural mechanisms and computational models of metacognition is crucial not only for understanding human cognitive processes but also for improving the resilience and flexibility of AI systems. Future studies in this field should expand the scope by integrating broader and more qualitative dimensions, such as affective self-assessment and social cognition, while maintaining the precision of current evaluation approaches.