Back to blogAI

AI Knows What We Say - But Not What We Know

Generative AI can access what we have documented, but tacit knowledge remains rooted in lived experience, judgment, and practical wisdom.

Table of Contents

Introduction

Generative AI systems are trained on vast corpora of human knowledge - books, articles, documentation, and digital communication. They represent an unprecedented aggregation of what humanity has been able to articulate.

Yet this raises a fundamental question:

Does access to everything we have written equate to access to everything we know?

The distinction between these two is not merely technical. It is philosophical - and long established.

The Greek Foundations of Knowledge

The roots of this distinction can be traced to Aristotle, who differentiated between multiple forms of knowledge:

  • Epistēmē (ἐπιστήμη): Theoretical, scientific knowledge concerned with universal truths
  • Technē (τέχνη): Craft or skill, involving production and replication
  • Phronesis (φρόνησις): Practical wisdom, the ability to act appropriately in context

While epistēmē and technē can be taught, documented, and transferred through instruction, phronesis resists formalisation.

It is:

  • situational
  • experiential
  • dependent on judgment

Phronesis is not reducible to rules. It requires discernment - an ability to navigate ambiguity and context. Like a witcher reading the signs of an unknown contract, you must learn to read the terrain itself, not just follow a map.

This form of knowledge is central to human decision-making, yet inherently difficult to codify.

Polanyi and the Structure of Tacit Knowledge

In the 20th century, Michael Polanyi reframed this philosophical distinction through the concept of tacit knowledge.

His assertion remains one of the most cited insights in knowledge theory: "We know more than we can tell."

Polanyi distinguished between:

  • Explicit knowledge: Formalised, codified, and communicable
  • Tacit knowledge: Personal, context-dependent, and often inexpressible

Tacit knowledge includes:

  • perceptual abilities (recognition, interpretation)
  • embodied skills (movement, coordination)
  • intuitive judgment (decision-making under uncertainty)

Importantly, tacit knowledge is not simply undocumented knowledge. It is knowledge that cannot be fully captured without losing its essence.

Tacit Knowledge in Organisations

In organisational contexts, tacit knowledge manifests in subtle but critical ways:

An experienced support engineer reads a ticket and knows something is wrong with the database before checking the logs. Not because they followed a diagnostic tree. Because they've seen this pattern before - the specific configuration of symptoms that points to one answer among many possibilities.

A project manager reads a status update and senses a team is in trouble, even though everything on paper looks fine. They're reading signals that aren't written down - the small hesitations, the gaps between what is said and what matters.

A senior designer knows when to break the rules, when the data is lying, when the metrics don't capture what the product actually needs to be. They've navigated enough strange territory to trust their instincts when the compass spins.

These are not edge cases. They are often the difference between success and failure.

And yet, they rarely exist in documentation.

The Implicit Boundary

Taken together, Aristotle's phronesis and Polanyi's tacit knowledge point to a consistent conclusion:

There exists a boundary between what can be articulated and what can only be experienced.

This boundary has historically shaped:

  • education (apprenticeship vs instruction)
  • organisations (mentorship vs documentation)
  • expertise (practice vs theory)

You cannot hand someone a map of a kingdom and expect them to understand how to move through it with grace. They must walk the paths themselves.

It now becomes highly relevant in the context of AI.

AI and the Domain of Explicit Knowledge

Modern AI systems are built on explicit knowledge at scale.

They process language, identify patterns in documentation, extract rules from written examples, and generate outputs based on what has been written and recorded. They excel at this. They can search, reorganise, synthesise, and retrieve explicit knowledge with remarkable efficiency.

Which leads to a critical observation:

AI systems are, by design, limited to the domain of explicit knowledge.

They cannot access what we know but cannot tell. They cannot learn from experience they have not had. They cannot develop intuitive judgment through practice. All the lore in the library does not grant you the sight.

What This Means

This is not a flaw in current AI systems. It is a fundamental architectural limitation.

AI trained on a trillion words still cannot know what experienced people know through years of working in a domain. It cannot replace judgment. It cannot replace the practical wisdom that comes from navigating ambiguity and context repeatedly.

For organisations, this has a direct implication:

If you believe AI has solved your knowledge problem, you have solved only the explicit half. The half that was already codifiable.

What remains is the tacit half. The judgment. The intuition. The practical wisdom that cannot be documented.

That still requires people. Still requires mentorship. Still requires time.

The Knowledge Manager's Reckoning

For those of us working in knowledge management, this distinction is foundational. We have always known:

  • Document what can be documented.
  • Protect and transfer what cannot be.

The error many organisations make is conflating the two. They attempt to document judgment. They try to codify intuition. They build decision trees for practical wisdom.

It doesn't work.

The emergence of AI as a tool for explicit knowledge should clarify - not obscure - this distinction.

Use AI to systematise what can be systematised. Free your experienced people to mentor, to transfer tacit knowledge, to develop judgment in others.

Do not use it as an excuse to eliminate the people who actually know things.

Conclusion

AI knows what we say. It has unprecedented access to explicit knowledge.

But it does not know what we know.

That knowledge - the phronesis of Aristotle, the tacit knowledge of Polanyi - remains irreducibly human.

The organisations that understand this distinction will thrive. Those that don't will end up with impressive documentation and junior people who keep repeating the same mistakes.

The boundary between explicit and tacit knowledge has existed for millennia. AI does not erase it.

It merely makes it more visible - and more important - than before.

References

  • Aristotle. *Nicomachean Ethics*. Translated by W.D. Ross. Oxford University Press, 1925.
  • Polanyi, Michael. *The Tacit Dimension*. University of Chicago Press, 1966.
  • Polanyi, Michael. *Personal Knowledge: Towards a Post-Critical Philosophy*. University of Chicago Press, 1962.