Metacognitive Skill Learning in Humans and AI | Stanford

1. Metacognitive Skill Defined, according to Smith

Timestamp: 00:01:02 to 00:02:55 - watch this moment on skim

Metacognition is the ability to monitor and control one's own mental states, including attention, emotions, and memory. It is crucial for both human and AI cognitive performance, and surprisingly, it is an even better predictor of learning and performance outcomes than IQ. Those with strong metacognitive skills can outperform those with higher IQs but lower metacognitive abilities, making it a vital 21st-century competency.

Significance (High): Highlights the importance of metacognition in learning and performance, suggesting that it can compensate for lower IQ.

Sources in support: Brandon Conan Smith (Speaker)

2. Smith on the Lack of Formal Theory

Timestamp: 00:03:20 to 00:05:08 - watch this moment on skim

Metacognitive skill is not well understood, and there is no formal theory of how the cognitive mechanisms underlying it generate the phenomena we perceive. This lack of a formal theory limits its application to both humans and AI, as humans have inbuilt metacognitive skill learning mechanisms, but AI does not. Therefore, a unifying theory of metacognition is needed to integrate research, distinguish it from general cognition, inform applications, and clarify how improvements in monitoring and control occur. Smith aims to answer Shraw's call for a unifying theory of metacognition.

Significance (Medium): Emphasizes the need for a formal theory of metacognition to advance its application in both humans and AI.

Sources in support: Brandon Conan Smith (Speaker)

3. Smith Explains Dual System Metacognition

Timestamp: 00:06:34 to 00:08:15 - watch this moment on skim

There are two basic forms of metacognitive information types: type one, which includes affective, non-propositional metadata and metacognitive feelings that are fast and automatic, such as feelings of knowing or rightness; and type two, which includes metacognitive strategies that are slower, declarative, and propositionally structured, such as concepts and strategies that allow us to direct our processes towards retrieving information. Type one metacognitive feelings can trigger type two metacognitive strategies, allowing us to retrieve knowledge that would otherwise remain inaccessible.

Significance (Medium): Explains the two types of metacognition and how they interact to retrieve knowledge.

Sources in support: Brandon Conan Smith (Speaker)

4. Smith on the Characteristics of Skill

Timestamp: 00:09:44 to 00:11:43 - watch this moment on skim

Metacognitive skill embodies the same characteristics as motor and cognitive skill, including goal structure and knowledge types. Complex goals require sub-goals, entailing a hierarchical goal structure. Knowledge allows an agent to choose the right actions to achieve goals, directed by declarative knowledge (propositional facts, rules, explicit reasoning) and procedural knowledge (implicit representations that execute the control of the action). Procedural knowledge builds up over time through practice, becoming fast and automatic, eventually replacing declarative knowledge.

Significance (Medium): Highlights the common characteristics shared between metacognitive, motor, and cognitive skills.

Sources in support: Brandon Conan Smith (Speaker), Joshua Shepard (Researcher), Nelson Narin (Researcher)

5. Smith Details Proceduralization Theory

Timestamp: 00:13:28 to 00:15:22 - watch this moment on skim

The theory of metacognitive skill learning relies on a theory of proceduralization, where slow declarative knowledge is converted into fast procedural knowledge that is increasingly refined. Declarative knowledge moves into working memory, allowing you to activate actions, while procedural knowledge operates outside working memory. Over practice, procedural knowledge builds up to become faster, automatic, and eventually replace declarative knowledge, allowing for cognitive reinvestment where working memory is freed up for higher-level control. This allows you to apply your working memory to monitor the situation for changing circumstances, which allows you to apply your skills more flexibly.

Significance (High): Explains how declarative knowledge is converted into procedural knowledge, freeing up working memory for higher-level control.

Sources in support: Brandon Conan Smith (Speaker)

6. Smith on Proceduralization Hallmarks

Timestamp: 00:16:42 to 00:18:35 - watch this moment on skim

The hallmark signs of proceduralization seen in motor and cognitive skill should also be seen in metacognitive skill, specifically the power law function, which quantifies the speeding up of reaction times. Attentional skill and metamemory skill also follow a power law of learning, reflecting the characteristics of motor and cognitive skill. This theory also helps to understand confounding data in empirical research, such as why athletes who are regularly not very self-conscious are more likely to have their performance disrupted under pressure.

Significance (Medium): Connects proceduralization to the power law of learning and explains confounding data in skill research.

Sources in support: Brandon Conan Smith (Speaker), Logan (Researcher)

7. Smith on Metacognitive Proceduralization

Timestamp: 00:20:06 to 00:21:53 - watch this moment on skim

Metacognitive proceduralization has been unidentified because it is invisible to observers and less perceivable to the performers themselves. Procedural knowledge is not conscious and operates outside of working memory, so when skill develops, it becomes an unconscious habit. As meta-knowledge is converted into imperceptible procedural knowledge, it disappears below the horizon, and people lose conscious access to the strategies they use to monitor and control their attention and learning strategies. This dual imperceptibility explains why metacognitive proceduralization has been unknown to date.

Significance (Medium): Explains why metacognitive proceduralization has been difficult to identify due to its imperceptibility.

Sources in support: Brandon Conan Smith (Speaker)

8. Smith on Human Computer Interaction

Timestamp: 00:22:01 to 00:23:24 - watch this moment on skim

The speaker discusses human-computer interaction (HCI) and how metacognition isn't classically applied. The theory helps reveal where metacognition fits into HCI and how we can better apply it. The metahci interface improves the classic GMS model by adding metacognitive processes like self-monitoring, self-regulation, and adaptive reflection. It helps reveal where users apply metacognitive skill and how we can better manage and train them to engage in complex tasks and decisions better.

Significance (Medium): Explains how metacognition can be integrated into human-computer interaction to improve user performance.

Sources in support: Brandon Conan Smith (Speaker)

9. Smith on AI's Metacognitive Struggles

Timestamp: 00:24:04 to 00:25:44 - watch this moment on skim

AI has been making incredible advancements, but it struggles the most in the metacognitive domain. There has been a trajectory towards increasing self-reference in AI design, with processes becoming increasingly self-monitoring, self-controlling, and self-learning. AI developers have noticed that using metacognition can significantly improve performance, such as in Metarag, which improves its accuracy by detecting and fixing reasoning errors, and Quietstar, which uses an inner monologue to consider the best option before outputting the answer.

Significance (High): Highlights the importance of metacognition in AI and provides examples of how it can improve performance.

Sources in support: Brandon Conan Smith (Speaker), Boston Dynamics (Organization), Google (Organization), Metar (Organization)

10. Smith on the Limitations of AI

Timestamp: 00:25:44 to 00:27:34 - watch this moment on skim

Many problems in AI result from systems having poor metacognition, including a lack of robustness, an inability to be flexible or deal with unpredictable environments, struggles with ambiguity and uncertainty, and challenges for cooperation and safety. These limitations lead to ethical concerns, as AI systems may be misaligned with human values and misinterpret human intentions. Importing insights into human metacognitive skill learning can be very useful, particularly the theory of metacognitive proceduralization, which would allow systems to automatize their strategy use and adapt over time.

Significance (High): Explains the limitations of AI due to poor metacognition and how importing human metacognitive skills can improve AI systems.

Sources in support: Brandon Conan Smith (Speaker), Johnson (Researcher)

11. Smith on Metalearning Reward Function

Timestamp: 00:29:07 to 00:30:49 - watch this moment on skim

At triple AI, Smith proposed a metalearning reward function where metacognitive strategies would become automated as they are used and rewarded over time. The artificial system would identify a learning task, select the appropriate learning strategy, apply the strategy, and evaluate its effectiveness. If useful, it would result in a positive reward; if not, a negative reward. These rewards and punishments would feed back into its memory of strategies, where useful strategies would become more automated with time, allowing for more adaptive improvements under uncertainty.

Significance (Medium): Describes a metalearning reward function that can help AI systems automate and improve their learning strategies.

Sources in support: Brandon Conan Smith (Speaker)

AI Advocacy (Low severity)

Type: Professional

Brandon Conan Smith's research focuses on improving AI through metacognitive skills. This raises questions about whether his enthusiasm for AI could color his perception of its potential risks and ethical implications.

Significance: The audience is left to wonder if Smith's advocacy for AI might lead to an underestimation of the challenges and potential downsides associated with advanced AI systems.