Metacognition as Framework for AI Self-Governance

A recent position paper available on arXiv suggests that metacognition should be the foundational scientific approach for effective self-governance in generative AI. The authors argue that AI systems must manage their generative processes while navigating uncertainty and lacking context. They advocate for metacognitive alignment at various levels: computational, algorithmic, and ecological. At the computational tier, metacognition encompasses functions such as monitoring, evaluation, control, and adaptation. Meanwhile, at the algorithmic level, these functions are executed through methods like elicitation, iteration, and modularization. This paper is cataloged as arXiv:2605.23981v1.