Metacognition Should Be the Scientific Framework for Bounded and Effective Self-Governance in Generative AI
A new position paper proposes metacognition as the foundational scientific framework for generative AI self-governance. The paper argues that for AI systems to effectively manage their own generative processes, especially in uncertain or data-scarce situations, they require metacognitive capabilities. This approach aims to align computational, algorithmic, and ecological levels to ensure AI is both capable and well-governed, rather than viewing these as conflicting goals. AI
IMPACT Proposes a new theoretical approach to ensure AI systems are both capable and safely governed.