Indwelling

In the AI Story

Polanyi developed indwelling as part of his larger phenomenological account of knowing. He observed that all perception, all skilled performance, and all understanding involve attending from clues we cannot fully specify to meanings that emerge through their integration. The blind person's cane became his canonical example because it made visible a structure that operates in all tool use: the instrument that has been successfully indwelt becomes part of the perceptual system rather than an object in the world. The cane's tip becomes, phenomenologically, the boundary of the body—the point where sensation begins. This is not metaphor but description of lived experience.

The concept illuminates both the exhilaration and the risk of AI tool adoption with unusual precision. When builders describe the experience of working with Claude Code—the sense that the tool has become an extension of thought, that removing it would feel like losing a cognitive faculty—they are describing successful indwelling. The tool has been incorporated into the builder's from-to structure: she attends from the tool's capabilities to the creative work she is pursuing, and the tool's mediation has become transparent. This transparency is what enables flow, what produces the twenty-fold productivity multipliers, what makes the imagination-to-artifact ratio collapse. The tool works because it has been indwelt.

But indwelling carries an intrinsic risk that mechanical tools do not pose at the same intensity. When a tool is indwelt, critical evaluation of the tool's reliability is suspended—the from-to structure requires trust in the subsidiary elements. The pianist trusts her piano. The surgeon trusts her scalpel. The trust is well-founded because these tools are mechanically predictable. But when the indwelt tool is a large language model—a system capable of producing outputs that are confidently wrong, that hallucinates with the same fluency it reasons—the trust that indwelling requires can be catastrophically misplaced. The tool produces an output that meets every surface criterion of quality. The builder, attending from the output rather than to it, integrates it into her focal work. The wrongness remains undetected because the from-to structure that enables productive flow is the same structure that suppresses the scrutiny that would catch the error.

The discipline required for safe indwelling of unreliable tools is the deliberate oscillation between transparency and opacity—between the mode in which the tool is subsidiary (enabling flow) and the mode in which the tool becomes focal (enabling evaluation). This oscillation is cognitively expensive and phenomenologically disruptive. It breaks flow. But it is necessary. The builder must periodically refuse to trust the tool—must shift from attending through it to attending to it, subjecting its outputs to the focal scrutiny that indwelling normally suppresses. This oscillation cannot be automated or delegated. It is the irreducible human contribution in any collaboration with a tool capable of sophisticated error.

Origin

Polanyi developed the concept in the 1950s and gave it its systematic treatment in The Tacit Dimension (1966). The term "indwelling" was chosen deliberately to carry its theological and mystical resonances—the idea of dwelling within something, being at home in it—while applying those resonances to the prosaic reality of tool use. Polanyi wanted to dissolve the Cartesian separation between subject and object, mind and world, by showing that skilled engagement involves a kind of inhabitation: the tool-user dwells in the tool, and through that dwelling, the world the tool reveals becomes directly present to consciousness.

Key Ideas

Tools become senses. Successful indwelling transforms an instrument into a perceptual organ—the cane becomes tactile extension, the AI becomes cognitive extension, each revealing aspects of reality invisible without the tool.

Transparency enables performance. The tool must disappear from focal awareness for skilled use to occur—conscious attention to the instrument disrupts the from-to structure and collapses performance.

Trust is structurally required. Indwelling depends on trust in the subsidiary elements—the pianist must trust the keys will respond, the builder must trust the code will execute—making critical evaluation phenomenologically difficult during flow.

Unreliable tools pose novel danger. When the indwelt tool can produce confident errors indistinguishable from competent outputs, the trust that enables indwelling becomes a vector for systematic failure.

Oscillation is the discipline. Safe use of powerful AI requires deliberate movement between trusting transparency (flow) and critical opacity (evaluation)—a rhythm that must be consciously maintained against the tool's seductive convenience.