Materials That Talk Back

In The You On AI Field Guide

Ingold documented this across cultures with ethnographic precision. Pacific Island knotters describe cordage as having a will — a tendency to twist that must be respected, not overcome. Scottish drystone wallers speak of each stone having a face, a preferred orientation the mason discovers through handling. Every ceramic tradition knows clay talks back through moisture, particle size, mineral composition, and temperature history — the potter must read and respond in real time, adjusting technique in continuous dialogue. This is not charming animism but accurate phenomenology of skilled practice: materials do contribute, and the contribution is not reducible to their documented properties. A database under load behaves in ways no manual predicted; the migration fails, and the failure is the computational material talking back, revealing something the specification did not contain.

The talking back is epistemically productive. When the birch splits unexpectedly, the carver encounters the world's independent behavior — something unpredictable from surface inspection and undocumented in manuals. This surprise forces learning about birch that cannot come from descriptions alone. The trace deposited by this encounter becomes part of enacted knowledge, available for every future engagement with similar material. In contrast, when Claude produces an unexpected connection — linking adoption curves to punctuated equilibrium in Segal's account — the surprise comes from linguistic association, not material encounter. It enriches representational knowledge (extending the map of concept-connections) but does not deposit the kind of trace that material resistance produces. Both surprises are valuable; they contribute differently.

The computational substrate has its own logic and talks back — but increasingly, the programmer's encounter with that logic is mediated by AI. In the old workflow, code fails, the programmer debugs, the system's behavior reveals a discrepancy between intention and execution. This is material talking back. In the new workflow, Claude generates code that works; if it fails, the programmer describes the symptom to Claude, which generates a fix. The computational material still has independent behavior, but the encounter with it is mediated by a system that interprets rather than resists. The programmer describes; the model remedies. The enacted knowledge that debugging deposits — the specific, bodily understanding of how this system behaves under these conditions — is not produced because the material encounter did not occur.

Origin

The concept emerges from Ingold's synthesis of Gibson's affordance theory with phenomenological accounts of tool use and anthropological fieldwork. The Perception of the Environment (2000) introduced the idea that environments are not passive stages but active participants in the lives of organisms. Making (2013) extended this to materials in craft practice. The language of 'talking back' appears throughout Ingold's work as a deliberate provocation against the hylomorphic model's assumption of passive matter. Philosophically, it draws on Heidegger's analysis of the tool's withdrawal and Merleau-Ponty's reversibility — the chiasm in which touching is simultaneously being touched. Empirically, it is grounded in the testimony of practitioners across cultures who describe materials as contributors, not substrates.

Key Ideas

Materials actively contribute information. Through resistance, deflection, and unexpected behavior, materials provide feedback that shapes the maker's decisions and educates perceptual capacity.

Talking back is not metaphorical. It is structural description of the epistemic feedback loop in skilled practice — the mechanism through which enacted knowledge is deposited trace by trace.

Material surprise differs from associative surprise. Encounter with wood's unexpected behavior forces learning from the world; AI's unexpected connections draw from linguistic patterns — both valuable, epistemically different.

Mediated encounter is not material encounter. When programmers describe symptoms to AI rather than debugging directly, the computational material's 'talk back' is filtered through a representational system — the trace that manual debugging would deposit is not laid down.