We Know More Than We Can Tell

The idea in plain words

You can recognise a friend’s face among thousands, but you cannot describe it well enough for a stranger to do the same. A good doctor senses that a patient is more ill than the tests suggest. An experienced engineer looks at a design and feels that something is wrong before finding it. In each case the knowledge is real and reliable, and it cannot be fully stated.

That is tacit knowledge: what people know how to do, or see, without being able to say exactly how. It is learned through practice, example and apprenticeship, not from manuals, and it is lost when the people who hold it go.

Where it comes from

Michael Polanyi was a distinguished physical chemist who turned to philosophy in mid-career. In Personal Knowledge (1958) and The Tacit Dimension (1966) he argued against the ideal of fully explicit, impersonal knowledge. His summary was that “we can know more than we can tell.” Even science, he held, depends on scientists’ trained judgement about which problems are promising and which results are significant, a skill passed from master to apprentice.

In 1995 the management scholars Ikujiro Nonaka and Hirotaka Takeuchi built on Polanyi to explain how Japanese companies innovated: by converting tacit knowledge into explicit concepts and back again, through shared experience, metaphor and practice. The sociologist Harry Collins later distinguished several kinds of tacit knowledge, some of which can in principle be made explicit and some, embedded in social life, which cannot.

What it does to the mind

Tacit knowledge works through attention to particulars. Polanyi described it as attending from clues to what they mean, the way a reader attends from letters to sense. Experts notice patterns without being aware of the clues, which is why their explanations of their own skill are often wrong. It also explains a common frustration: experts asked to write down how they work produce rules that, followed by others, fail.

Where it shows up

The essay on large language models asks what happens when systems learn only from what was written down. The essay on enterprise software describes companies losing the unwritten knowledge of experienced staff during implementations, and sub-postmasters whose knowledge that they had not taken money counted for nothing against a system’s record. The essay on programming languages treats code as an attempt to make knowledge fully explicit.

What it asks of someone deciding

Before automating or documenting a task, ask the people who do it what they notice that the procedure does not mention. Their answers are where the risk lies.

Keep apprenticeship alive. Tacit knowledge passes through working alongside people, so structures that isolate juniors from experts, or remove the routine work through which skill is learned, erode it.

Respect unexplained judgement, but test it. Experts’ feelings can be right without explanation; they can also be habits. Track whether their intuitions are borne out.