Knowing-in-Action

In the AI Story

The philosophical foundation for knowing-in-action predates Schon by decades. Michael Polanyi's The Tacit Dimension (1966) articulated the claim that "we know more than we can tell," and that the most important knowledge — the knowledge that makes skilled performance possible — resists full articulation not as a practical matter but as a logical one. Gilbert Ryle's earlier distinction between knowing-that and knowing-how laid similar groundwork. Schon's contribution was to take these philosophical insights into the empirical study of professional practice, and to demonstrate that knowing-in-action is not marginal but central to what distinguishes the master from the competent.

The repertoire that enables knowing-in-action is built through what this volume calls geological understanding — the layered deposition of experience through thousands of hours of engagement with a medium that resists. The diagnostician who reads a chest X-ray in eleven seconds sees what the resident cannot see after eleven minutes, not because her eyes are sharper but because her repertoire is deeper. Each prior X-ray deposited a layer. The layers accumulated into a perceptual competence that operates below the threshold of articulation. Ask her how she saw the mass, and she will pause. The seeing was the knowing, and the knowing does not live in language.

The AI moment creates a specific asymmetry that Schon's framework illuminates with uncomfortable clarity. AI replicates knowing-that with extraordinary completeness — the articulable substrate of professional knowledge. It does not replicate knowing-how or knowing-in-action, because these forms of knowledge were never in text to begin with. The professional equivalent of the potter's competence is what the senior engineer in Trivandrum recognized as her twenty percent — the judgment, the instinct, the taste — that remained hers after Claude absorbed the other eighty. The eighty percent was the articulable. The twenty was the embodied.

The interaction between knowing-in-action and AI is more subtle than replacement. When the practitioner evaluates Claude's output, her evaluation is partly explicit (she reads the code, checks the logic) and partly tacit (she feels whether the architecture is right). The tacit evaluation is the most important contribution the practitioner makes to the partnership. It is also the most endangered. The tool's output is so polished that the practitioner's vague unease feels inadequate by comparison. The aesthetics of the smooth creates constant pressure to override the tacit verdict in favor of the visible, functional artifact. The author of The Orange Pill describes this dynamic in his account of the Deleuze passage: the code compiled, the prose flowed, and an overnight unease — knowing-in-action doing its work — was the only signal that something was wrong.

Origin

Schon introduced knowing-in-action in The Reflective Practitioner (1983) as the cognitive substrate beneath reflection-in-action, drawing explicitly on Polanyi's tacit dimension and implicitly on Ryle's earlier knowing-how/knowing-that distinction. He elaborated the concept in Educating the Reflective Practitioner (1987), where he argued that professional education's neglect of knowing-in-action was the deepest failing of the technical-rationality curriculum.

Key Ideas

Competence as performance. The knowing is the doing; the knowledge does not exist outside the skilled action that expresses it.

The tacit dimension. Polanyi's insight operationalized — the most important professional knowledge cannot be fully articulated because articulation presupposes it.

Built through friction. Knowing-in-action develops through specific kinds of resistance that force reorganization of perception itself.

AI replicates knowing-that, not knowing-in-action. The audit performed by large language models separates, with unprecedented precision, the articulable from the embodied.

The irreducible human contribution. The practitioner's tacit evaluation of AI output is what the partnership cannot function without, and what the partnership makes hardest to exercise.

Debates & Critiques

Skeptics argue that knowing-in-action is a temporary category — that sufficiently rich datasets and sufficiently sophisticated models will eventually articulate what currently resists articulation. Defenders respond that the argument misunderstands the concept: tacit knowledge is not difficult-to-articulate knowledge but constitutively non-propositional knowledge, and no amount of data will convert a potter's hands into a dataset. The empirical question of whether AI will eventually close this gap remains open; the conceptual question of what would count as closing it is where the debate actually lives.