AI can't have tacit knowledge—that's why it can't replace you
I shall reconsider human knowledge by starting from the fact that we can know more than we can tell.
— Michael Polanyi, The Tacit Dimension (1966)
The Great Automatic Grammatizator, a short story written by Roald Dahl in 1954, presents a fictitious world in which an inventor concludes that English grammar is merely a sequence of mathematical rules and proceeds to invent a machine that can "write" short stories. These machine-generated stories eventually get published in magazines and gain widespread acceptance, putting human authors out of work.
Seven decades after Dahl wrote this story, the parallels between its world and ours are hard to miss. In less than four years, AI went from being an amusing piece of software to a force that cost thousands of people their jobs. Layoffs attributed to AI in the U.S. across industries amount to over 150,000 between 2023 and May 2026. Among all industries, the technology sector is the largest job cutter, given the promise that AI can develop software, close deals, and talk to customers—all without human intervention.
On the flip side of this promise, companies are increasingly finding it difficult to extract returns from AI investments. Some firms see the cost of using AI exceed that of paying human employees, making it harder to justify the AI-driven layoffs. And 90% of companies seem to be in favor of increasing their AI budgets despite falling short of their estimated cost savings and not seeing immediate value.
Looking beyond the usual reasons for AI adoption concerning productivity and efficiency gains, a question remains: What made businesses confident that AI can automate knowledge work, which requires human intellect and understanding, in the first place? The answer might lie in how they understand organizational knowledge.
What can and can't be said
There are two kinds of organizational knowledge: explicit knowledge that can be articulated and documented and tacit knowledge that can't be articulated and can only be demonstrated.
Though it sounds abstract, it's visible in everyday work. When an experienced salesperson senses a deal is slipping even when the metrics look healthy, or a senior manager spots a flaw in an instant that junior colleagues missed for weeks, they're drawing on wisdom accumulated over years that they can't fully articulate when asked. This wisdom—comprising intuition, skill, judgment, and taste—is tacit knowledge.
Michael Polanyi, the theoretical physicist and philosopher who formulated the concept of tacit knowledge, argued that explicit knowledge depends on tacit knowledge, as the latter functions as the foundation of the former. "While tacit knowledge can be possessed by itself," he wrote, "explicit knowledge must rely on being tacitly understood and applied. Hence all knowledge is either tacit or rooted in tacit knowledge. A wholly explicit knowledge is unthinkable."
Expanding on this idea further, Ikujiro Nonaka, the Japanese organizational theorist, wrote in the Harvard Business Review that tacit knowledge and explicit knowledge interact inside an organization in four different ways:
From tacit to tacit: When one employee absorbs tacit knowledge from another through observation, imitation, and practice
From explicit to explicit: When an employee combines discrete pieces of explicit knowledge from different sources into a new whole
From tacit to explicit: When an employee articulates an innovative solution to a problem based on their accumulated experiences
From explicit to tacit: When employees internalize new explicit knowledge to broaden, extend, and reframe their own tacit knowledge
The knowledge-generating company, as Nonaka puts it, is one where these four ways co-exist in dynamic interaction to create a "spiral of knowledge". Both tacit and explicit knowledge is necessary for an organization—one can't exist without the other, as Polanyi theorized earlier.
While businesses in the West equate redundancy with inefficiency, he notes that successful Japanese companies deliberately assign redundant tasks to the workforce, like having multiple teams being tasked to come up with different approaches for the same project or rotating employees between different functions.
"Redundancy is important because it encourages frequent dialogue and communication," Nonaka explains. Employees often sharing overlapping information among themselves creates what he terms a "common cognitive ground" that facilitates the transfer of tacit knowledge that would otherwise remain unarticulated.
Why tacit knowledge eludes AI
For an AI to be perfectly capable of doing the tasks that humans have done so far, it needs both explicit and tacit knowledge. But large language models—the software foundation that powers generative AI—function through calculating statistical probabilities based on its massive trove of training data to answer queries. Training data, comprising only what has already been articulated and recorded, is explicit knowledge by definition. So AI is capable only at repurposing existing explicit knowledge.
Tacit knowledge is structurally incompatible with AI systems because it can't be fed as training data. It stems from lived experience, which a piece of software—no matter how powerful and autonomous—can never have. Without this lived experience, traits like situational awareness to spot hidden opportunities, creativity to tinker with different ideas, and decision-making based on real-world context and judgment will always be beyond AI's reach.
Yet when business leaders mandate the use of AI to automate entry-level tasks in the name of efficiency, it can result in the de-skilling of young professionals. The time-consuming and redundant aspects of such tasks are precisely what help develop judgment, context, and expertise, making these professionals better at their jobs over time.
The undermining of tacit knowledge can also have another unintended consequence: the stifling of innovation. It's the collective understanding that an organization possesses about customers, competitors, and market dynamics that will help in assessing any unprecedented situation and responding with an innovative solution. That understanding, which accumulates over time, is tacit and can't be separated from the people who hold it.
From the spinning jenny to modern software, every advancement in technology revolves around the promise of automating work that's manual, repetitive, and redundant. But the control still remained with humans, as these were tools that required someone to decide how they would be used.
However, the vendors selling AI make a claim that's unheard of in the history of technology: that, like humans, it can engage in reasoning and comprehension and even function autonomously. Such anthropomorphizing language veils the fact that AI is a probabilistic machine which functions solely on explicit knowledge. And explicit knowledge alone isn't enough to do work that requires creativity, context, and craftsmanship—all of which fall under the domain of tacit knowledge.
"By taking away the easy parts of his task, automation can make the difficult parts of the human operator's task more difficult," wrote cognitive psychologist Lisanne Bainbridge in her research paper Ironies of Automation. When businesses adopt AI aggressively without taking the importance of tacit knowledge into account, they can end up creating new problems rather than solving current ones.
After all, AI can tell a lot, but it's humans who really know.
