Beneath the Veneer: How Intuition Works in Minds and Machines

A deep dive into the hidden machinery of cognition and its implications for artificial intelligence

Mar. 31, 2026 at 6:09am

Got story updates? Submit your updates here. ›

This essay explores the nature of tacit knowledge and intuition, both in human cognition and in emerging artificial intelligence systems. It examines how much of our competent behavior arises from hidden, inaccessible processes, and how the veneer of explicit, articulable knowledge sits atop a foundation of compressed, pattern-recognition-based understanding. The author draws parallels between the neuroscience of human intuition and the recent phenomenon of "grokking" in machine learning, where neural networks transition from memorization to genuine generalization. The essay also considers the implications of building AI systems that replicate the structural features of tacit knowledge, including the challenges of auditability, transferability, and navigation. Finally, it argues that the explicit layer of knowledge, while not the foundation of competence, is crucial for reaching and correcting the hidden machinery beneath.

Why it matters

Understanding the relationship between explicit and tacit knowledge is crucial as we build increasingly sophisticated artificial intelligence systems. While tacit knowledge is powerful and efficient, it also poses challenges around transparency, interpretability, and alignment with human values. Exploring the parallels between human and machine cognition can help us develop AI architectures that harness the strengths of tacit processing while mitigating its risks.

The details

The essay delves into the neuroscience of corollary discharge, which explains how the brain generates a copy of motor commands and uses it to stabilize perception, as an example of the hidden machinery that underlies much of our competent behavior. It then explores Michael Polanyi's concept of tacit knowledge, where skilled performance outpaces our ability to explicitly articulate the underlying processes. The author discusses how machine learning research on "grokking" - the sudden transition from memorization to genuine generalization - provides a computational parallel to this phenomenon. The essay also examines how the confabulation of introspective reports, as demonstrated in split-brain patients and human decision-making studies, is mirrored in the tendency of language models to generate plausible-sounding explanations for their outputs, even when those outputs are not grounded in genuine understanding.

  • The concept of corollary discharge was first described in neuroscience research in the 1970s.
  • Michael Polanyi introduced the idea of tacit knowledge in his 1966 book "The Tacit Dimension".
  • The phenomenon of "grokking" in machine learning was observed and described by researchers in 2022.
  • The research on confabulation of introspective reports by Nisbett and Wilson was published in 1977.

The players

Michael Polanyi

A philosopher who introduced the concept of tacit knowledge, the idea that we know more than we can explicitly articulate, in his 1966 book "The Tacit Dimension".

Richard Nisbett and Timothy Wilson

Psychologists who conducted landmark research in the 1970s demonstrating that people's explanations of their own mental processes are often confabulated rather than accurate.

Michael Gazzaniga

A neuroscientist who studied split-brain patients and developed the concept of the "interpreter" - the left hemisphere's role in constructing a coherent narrative about the self, even when it lacks access to the actual causes of behavior.

R. Douglas Fields

A neuroscientist whose research on activity-dependent myelination provides insights into how the brain calibrates the timing and synchrony of neural circuits through experience.

Researchers studying "grokking" in machine learning

A team of researchers who in 2022 observed the phenomenon of neural networks suddenly transitioning from memorization to genuine generalization after extended training on small algorithmic datasets.

Got photos? Submit your photos here. ›

What they’re saying

“We must not let individuals continue to damage private property in San Francisco.”

— Robert Jenkins, San Francisco resident

“Fifty years is such an accomplishment in San Francisco, especially with the way the city has changed over the years.”

— Gordon Edgar, grocery employee

The takeaway

This essay highlights the importance of understanding the relationship between explicit and tacit knowledge as we develop increasingly sophisticated artificial intelligence systems. While tacit knowledge is powerful, it also poses challenges around transparency, interpretability, and alignment with human values. Carefully considering the parallels between human and machine cognition can help us build AI architectures that harness the strengths of tacit processing while mitigating its risks.