The Structure of Discovery (Polanyi)
Polanyi's account of discovery overturns the standard scientific-method narrative of hypothesis, test, and verification. Real discovery, he argued, begins not with a formulated hypothesis but with an intimation—a pre-articulate sense, often no more specific than "something is there," that guides the researcher toward one line of inquiry rather than another. The researcher commits to this intimation before the evidence justifies commitment, pursues it through uncertain territory, and only afterward constructs the explicit hypothesis that textbooks present as the starting point. This structure—intimation, commitment, pursuit, validation—cannot be reversed. The intimation must come first, and the intimation is irreducibly tacit: it arises from the researcher's embodied sensitivity to the domain's unresolved tensions, her fine sense of what is plausible, her accumulated understanding of where the existing frameworks strain against reality. AI systems do not intimate. They compute probable outputs from existing data. They can articulate hypotheses, test them, and evaluate results. But they cannot exercise the tacit judgment that selects which hypothesis is worth testing from the vast space of hypotheses that could be formulated—and that selection is where discovery actually occurs.
In the AI Story
Polanyi illustrated the discovery structure with examples from his own scientific career and from the history of science. Henri Poincaré's famous account of mathematical discovery—conscious preparation, unconscious incubation, sudden illumination, verification—maps directly onto Polanyi's intimation-commitment-pursuit-validation sequence. The preparation phase builds the tacit ground. The incubation phase operates on this ground beneath conscious awareness. The illumination is the moment when an intimation rises to conscious attention. And the verification is the explicit phase that textbooks mistake for the whole. But the verification phase is possible only because the prior phases—all tacit, all personal, all involving commitment to directions that could not yet be justified—created the conditions for the discovery to occur.
The structure reveals why AI-assisted research, while dramatically increasing the speed of literature review and hypothesis articulation, may actually slow the rate of genuine discovery. Discovery requires going beyond existing data—pursuing connections the data suggests but does not confirm, following intuitions that resist explicit justification, committing resources to directions that may prove fruitless. AI systems are anchored to their training data. They excel at producing outputs consistent with what is already known. They cannot intimate what is not yet known, because intimation requires the tacit sensitivity to a domain's unresolved tensions that only embodied expertise provides. The system can search the space of articulated hypotheses with extraordinary speed. It cannot exercise the fine sense of plausibility that tells the researcher which hypotheses are worth pursuing.
The builder who brings a "half-formed idea" to Claude—a pattern she feels but cannot specify—is exercising the intimation phase of discovery. The AI's response—concrete, structured, drawn from vast associative range—accelerates the articulation phase. The collaboration is genuinely productive when the builder possesses the tacit ground to evaluate whether the articulation captures the intimation's truth or merely gives it plausible form. The collaboration fails when the builder accepts the articulation because it is smooth rather than because it is right—when the tool's surface competence substitutes for the evaluative judgment that only personal knowledge can provide. The Deleuze fabrication is discovery structure breaking down: elegant articulation accepted without the committed evaluation that would recognize it as sophisticated nonsense.
Origin
Polanyi developed the discovery framework across multiple works—Personal Knowledge (1958), The Tacit Dimension (1966), and his essays on the logic of tacit inference. The account drew on his experience as a physical chemist pursuing problems that formal methodology could not generate: How did he decide which lines of inquiry to follow? Why did some hunches prove fruitful while others, equally plausible at the start, led nowhere? The answer, Polanyi concluded, was that discovery depends on tacit intimations—pre-verbal senses of promising directions—that cannot be derived from explicit procedures or formalized into searchable algorithms.
Key Ideas
Intimation precedes hypothesis. Discovery begins with inarticulate sense of a hidden pattern—not a formulated claim but a felt direction the researcher commits to pursuing before justification is available.
Commitment under uncertainty. The researcher stakes time, resources, and reputation on intimations that may prove fruitless—this risk-taking is constitutive of discovery, not a regrettable necessity.
Fine sense of plausibility. The tacit capacity to distinguish promising from unpromising directions before evidence is available—built through years of immersion, irreducible to computation.
AI cannot intimate. Pattern-matching systems produce outputs consistent with training data but cannot sense where existing frameworks fail or generate the pre-articulate hunches that drive genuine discovery.
Articulation is late stage. The hypothesis-test-verify sequence that textbooks present as scientific method is the explicit phase of discovery—possible only after tacit intimation has selected what to make explicit.
Appears in the Orange Pill Cycle
Further reading
- Michael Polanyi, "The Logic of Tacit Inference," Philosophy (1966)
- Michael Polanyi, Personal Knowledge, Part Three (1958)
- Henri Poincaré, "Mathematical Creation" (1908)
- Thomas Kuhn, The Structure of Scientific Revolutions (1962)
- Karl Popper, The Logic of Scientific Discovery (1934)