Metacognitive Skill Learning in Humans and AI | Stanford

About this Video

• Video Title: Metacognitive Skill Learning in Humans and AI | Stanford
• Channel: Meta-Think
• Speakers: Brandon Conan Smith
• Duration: 00:31:37

Introduction

This Stanford talk by Brandon Conan Smith explores metacognitive skill learning in humans and AI. The presentation aims to explain the cognitive mechanisms underlying metacognitive skills in humans, outlining how this understanding can be applied to improve AI design and unlock its potential. Smith focuses on developing a formal theory of metacognitive skill learning, bridging the gap between psychological research and AI development.

Key Takeaways

1. Metacognition's Importance: Metacognitive skills—the ability to monitor and control one's mental states—are crucial for both human and AI cognitive performance, exceeding IQ as a predictor of success.
2. Formal Theory of Metacognitive Skill Learning: Smith presents a formal theory of metacognitive skill learning, integrating research and clarifying how improvements in monitoring and control occur. This theory is empirically validated.
3. Dual-System Metacognition: Metacognitive information processing involves two systems: fast, automatic, non-propositional "feelings" (e.g., feelings of knowing) and slower, declarative, propositional strategies for knowledge retrieval.
4. Proceduralization in Skill Learning: Skill learning across motor, cognitive, and metacognitive domains involves proceduralization—the conversion of declarative knowledge into fast, automatic procedural knowledge, leading to improved speed, accuracy, and flexibility. This frees up working memory for higher-level control.
5. AI Applications: The lack of inherent metacognitive abilities in AI is a major limitation. Smith proposes importing human metacognitive insights into AI design to improve robustness, flexibility, explainability, cooperation, and safety. A metalearning reward function is suggested to automate metacognitive strategies.