Updated: May 20, 2026

Key Points

• AI can widen the search space, but the human with tacit knowledge narrows the problem space.

• The value of the human with tacit knowledge comes through tacit judgment, tacit taste, and tacit strategy.

• Until the learning space is incomplete for the tacit knowledge, AI captures the statistical shadow of tacit knowledge, not tacit knowledge itself.

Introduction

More frequently, academic discussions revolve around the value of academics in a world where AI can generate text better and quicker than human authors. That’s right. AI can write fast. It can summarize, synthesize, and recombine ideas at a speed that no human author can match. But oftentimes, AI-generated text reads like it is talking in circles.

This happens because AI has traded productivity for directionality. It can produce more text, but it does not always know where the argument should go. AI does not have the sense of Occam’s razor, or at least not unless it is specified in the prompt. It does not naturally know when an argument has become sufficient, when the point has been made, or when it is time to stop and draw a conclusion. This knowledge of when the argument is “enough” is part of the tacit knowledge of the researcher.

Below, I break down the human contribution with the tacit knowledge into three components: judgement, taste, and strategy.

Tacit Judgment

Tacit knowledge includes the unspoken judgment behind the statement: why this argument, why now, why this framing, why this level of detail, and why this conclusion is sufficient. What AI frequently misses is not the ability to generate more material, but the ability to capture this tacit judgment during its training. AI can widen the search space, but the researcher narrows the problem space.

Of course, one could argue that humans also perceive tacit knowledge through pattern recognition. If AI mimics this property of the human nervous system, then perhaps it should be able to capture tacit knowledge as well. This argument is not entirely wrong. Tacitness is a matter of degree. Some tacit knowledge does leave traces in repeated behavior, expert writing, professional routines, and accumulated examples. To the extent that these traces are observable, AI can approximate them.

But the deeper problem remains: some forms of expertise are not stored as information in the first place. They are stored as judgment, attention, taste, professional intuition, and embodied experience. Knowing when an argument is “enough” is rarely written down as a rule. It is learned through practice, failure, revision, feedback, and repeated participation in a professional community. AI can imitate the textual residue of this process, but imitation of the output is not the same as possession of the underlying judgment.

To prove the point, in my conversation with two chatbots about their opinion on how well generative AI chats can capture tacit knowledge, they both gave a modest assessment of their ability. One of them even rated its ability to capture and reflect on tacit knowledge as low as 10%.

"The frontier models are genuinely impressive at the linguistic shadow of tacit knowledge. But the shadow is not the thing." (Sonnet 4.6 by Claude, 2026)

Tacit Taste

Creativity is often described as the recombination of known ideas. But in research, creativity is not just recombination. It is selective recombination. The creative act is not only to put existing ideas together, but to know which ideas belong together, which connection is meaningful, and which combination opens a new perspective.

This is where tacit knowledge matters. Creativity is in the recombination of known ideas with the tacit knowledge of a new perspective. AI can combine many ideas, but the researcher decides which combination is meaningful, which analogy is productive, which theory clarifies rather than decorates, and which question is worth pursuing. This is where the tasit taste comes through.

If tacit knowledge is absent, the purpose of recombination is absent. The text may still look rich, polished, and full of ingredients, but it starts to resemble a kitchen sink cookie: technically delicious, but without a memorable taste.

Strategic Vector

The third component of the value of the researcher is the strategic vector of what they are doing. Strategy-building also relies on tacit knowledge. It is not only about knowing what has been said, but also about knowing where the conversation should go next.

This is why AI can be useful but still incomplete. AI can generate options, summarize literatures, and suggest possible directions. But the researcher provides the evaluative vector. The researcher knows which direction is promising, which path is theoretically interesting, which simplification is useful, and which problem is worth solving now.

The strategic vector is what turns output into contribution. Without it, text may be coherent but not consequential. It may be informative but not directed. It may sound academic but still lack a clear intellectual destination.

Conclusion

The value of humans is in their ability to capture tacit knowledge: know-how that is realized through professional practice and resides outside of the fully codified knowledge of the profession. AI can approximate some traces of this knowledge, especially when they are repeated, documented, or visible in prior outputs. But the deeper layer of tacit knowledge remains tied to practice, judgment, attention, taste, and professional intuition.

The action of practicing your craft builds tacit knowledge and gives us the competitive advantage over AI. Use AI, but know what you bring to the table: your selective judgment, your creative taste, and your startegic vector.

References:

1. Nonaka, I. (1994). A dynamic theory of organizational knowledge creation. Organization Science, 5(1), 14–37.

2. Polanyi, M. (1966). The tacit dimension. Doubleday & Company, Inc.: Garden City, New York.

Please cite this article as: 

Petryk, M. (2026, May 20). Tacit Knowledge and AI: Keeping the Human Edge in the AI Age. MariiaPetryk.com. https://www.mariiapetryk.com/blog/post-32