The Friction That Produces Understanding

In the AI Story

The concept draws on the phenomenological tradition's analysis of how understanding is produced through engagement with resistant materials. The mechanic's hypothesis about the failing bearing is tested against the motorcycle's behavior; when the hypothesis is refuted, she must revise her understanding, and the revision deposits the thin layer of calibrated judgment that accumulates into expertise. The resistance is not incidental to the learning — it is the mechanism through which learning occurs. Bypass the resistance and you bypass the learning.

Edo Segal's ascending friction framework in The Orange Pill extends this analysis by proposing that AI relocates difficulty to higher cognitive floors rather than eliminating it. Crawford's framework is friendlier to this extension than it might appear. He accepts that some friction should be eliminated (mechanical tedium) and that some new difficulties emerge at higher levels (architectural judgment, evaluation). What Crawford adds is the empirical observation that the ascent is not automatic — the lower-level friction that produced lower-level understanding is not seamlessly replaced by higher-level friction producing equivalent understanding. Much of the lower-level engagement was itself the mechanism through which higher-level judgment developed. Removing the lower level may weaken rather than liberate the higher.

The distinction has direct implications for how AI tools should be deployed in professional formation. Junior practitioners in particular face the question of whether AI-mediated workflows will produce the embodied understanding that mid-career judgment requires. Crawford's framework predicts they will not — not because AI is inherently harmful but because the specific cognitive encounters through which understanding is built are the encounters AI is designed to eliminate. The prescription is not to refuse AI but to preserve, through deliberate institutional design, the occasions for productive friction that professional formation requires.

The concept also illuminates why AI-generated output can be competent without being understood. The output works because it passes the functional tests defined in training data. The understanding that would have enabled the practitioner to evaluate whether the output is genuinely appropriate — not merely functional — was deposited by friction she has not experienced. She becomes competent at direction without becoming competent at judgment, and the gap is invisible until a situation arises that direction alone cannot handle.

Origin

Crawford developed the concept across his corpus, with particular attention in Shop Class as Soulcraft (2009) and in essays on AI. The precise phrase is drawn from his engagement with the phenomenological tradition on resistant materials as epistemic conditions.

Key Ideas

Not all friction is equivalent. Mechanical tedium differs from productive friction; the distinction is invisible to productivity metrics that treat friction as uniformly negative.

Resistance as learning mechanism. The material's refusal to behave as expected is the specific occasion through which the practitioner's understanding deepens.

Geological deposition. Understanding accumulates in layers, each deposited by a specific productive encounter; bypass the encounter and no layer is deposited.

The ascent is not automatic. Higher-level friction does not substitute directly for lower-level friction; some of the higher capacities were produced by the lower engagement.

Deliberate preservation. Institutional structures must protect the occasions for productive friction that AI-mediated workflows systematically eliminate.