Why AI Can’t Create Knowledge
A 30-year-old framework explains modern AI’s most fundamental limitation
In 1994, Ikujiro Nonaka published a paper that would become one of the most cited works in organizational theory. “A Dynamic Theory of Organizational Knowledge Creation” introduced concepts - tacit and explicit knowledge, the SECI model, the knowledge spiral - that shaped how we think about learning organizations.
Three decades later, his framework explains something Nonaka never anticipated: why AI systems, despite their impressive capabilities, cannot create genuinely new organizational knowledge.
The Tacit-Explicit Distinction
Nonaka built on philosopher Michael Polanyi’s insight: “We can know more than we can tell.” Knowledge comes in two forms.
Explicit knowledge can be codified and transmitted in formal language - formulas, procedures, specifications. It lives in documents and databases.
Tacit knowledge is personal, context-specific, and hard to formalize. It lives in bodies and habits. A master craftsman has tacit knowledge in their fingertips. An experienced salesperson has tacit knowledge about reading customer reactions. This knowledge can be demonstrated but not fully explained.
Four Modes of Knowledge Creation
Nonaka’s central contribution was mapping how knowledge converts between tacit and explicit forms. He identified four modes:
Socialization creates tacit knowledge through shared experience. Apprentices learn from masters not through manuals but through observation, imitation, and practice. The key requirement is physical co-presence - shared experience.
Externalization articulates tacit knowledge into explicit concepts. This is the hardest mode and the source of genuine innovation. It happens through dialogue and, crucially, through metaphor - relating contradictory concepts to express what we know but cannot directly say.
Combination reconfigures explicit knowledge through sorting, adding, and recategorizing. This is where databases, reports, and information systems operate.
Internalization embodies explicit knowledge into tacit knowledge through practice - “learning by doing.” Reading a manual doesn’t create skill; applying it repeatedly does.
Where AI Lives in This Framework
Here’s what becomes clear when you apply Nonaka’s framework to AI systems: they operate almost exclusively in Combination mode.
Large language models excel at reconfiguring explicit knowledge. They sort, synthesize, recontextualize vast bodies of text. They can write documentation, generate summaries, create reports that combine information from multiple sources. This is genuinely useful - Combination is how knowledge spreads across organizational boundaries.
But consider the other three modes:
Socialization requires shared physical experience. AI has no body, no presence in the world, no capacity for the kind of cognitive synchronization that happens when humans work side by side. No amount of prompt engineering creates this.
Externalization requires tacit knowledge to articulate. AI can help humans express what they know - acting as a dialogue partner - but it has no tacit knowledge of its own. It can generate metaphors, but not from lived experience. The metaphors are combinations of patterns in training data, not bridges from embodied understanding to explicit form.
Internalization requires embodiment and practice. AI has no “doing.” There is no tacit layer for explicit knowledge to sink into, no habits formed through repetition, no bodily knowing that develops through action.
The Strategic Implications
This isn’t a critique - it’s a clarification. Understanding where AI operates in the knowledge creation cycle helps us deploy it appropriately.
AI augments Combination. It accelerates the reconfiguring, synthesizing, and recontextualizing of what has already been made explicit. Organizations have always had Combination bottlenecks - too much information to process, too many sources to integrate. AI relieves this pressure.
AI cannot replace the human-dependent modes. The tacit knowledge that makes organizations distinctive - the judgment, intuition, craft knowledge that competitive advantage often rests on - still requires human processes to develop and share.
The premium on Socialization and Externalization increases. If Combination gets automated, the scarce resource shifts to the modes AI can’t perform. The ability to create shared tacit understanding (Socialization) and to articulate tacit knowledge into shareable form (Externalization) becomes more valuable, not less.
This aligns with what we’re seeing empirically. Recent research on how workers want AI deployed shows consistent patterns: workers want AI for information-processing tasks and reject it for interpersonal and creative work. Through Nonaka’s lens, they’re saying: automate Combination, preserve what requires human presence.
What This Means for AI Strategy
Organizations investing in AI should ask: Which of our knowledge processes are we automating?
If the answer is primarily Combination - synthesizing documents, generating reports, processing information - AI can help significantly.
If the goal is genuine knowledge creation - developing new insights, building organizational capabilities, creating competitive advantage - AI is a tool, not a solution. The human processes that generate tacit knowledge and articulate it into shareable form remain essential.
The 1994 paper included a remarkable observation: “The importance of knowledge related to action has been recognized in the area of artificial intelligence.” Nonaka was referring to expert systems and early AI research. But his framework anticipated a fundamental truth that became even more relevant thirty years later.
Knowledge creation is not information processing. It requires continuous dialogue between what can be said and what can only be shown, between the explicit systems that scale and the tacit understanding that makes those systems meaningful.
AI has transformed what’s possible in Combination mode. But three of four modes still require what they always required: human presence, shared experience, and the embodied knowledge that we can know but never fully tell.
What knowledge in your organization lives in bodies and habits rather than documents and databases? What would change if you optimized AI for Combination while investing more deeply in Socialization and Externalization?
Related: 06-molecule—tacit-vs-explicit-knowledge, 06-atom—tacit-knowledge, 00-source—nonaka-1995-knowledge