Acquisition vs. Learning

In the AI Story

Gee drew the distinction from applied linguistics, where it had been articulated by Stephen Krashen and others in the context of second-language learning. Gee generalized the distinction to all forms of complex competence. The senior physician has acquired the Discourse of medicine through decades of immersive practice — she thinks like a physician without having to consciously apply rules. The medical student learning to make diagnoses is still at the learning stage — applying rules consciously, checking against textbook examples, confident only in the cases that closely resemble the training cases.

Both stages are necessary. Every acquired Discourse begins with learned elements — no one is born speaking their native language. But the transition from learning to acquisition requires the thousands of hours of practice through which explicit rules become tacit dispositions, through which effortful application becomes automatic response, through which the competence shifts from something the practitioner has to something the practitioner is.

AI threatens to extend the learning stage indefinitely by providing external scaffolding that substitutes for the internalization process. The practitioner using Claude can apply the rules of competent practice through the tool without having internalized the rules herself. The output looks like the output of an acquired practitioner. The practitioner's relationship to the output is closer to that of a learned practitioner — conscious, explicit, dependent on external reference. Under normal conditions, the dependence is invisible because the tool is available. Under conditions where the tool is unavailable, unreliable, or confronting novel problems it cannot handle, the practitioner discovers that the acquisition never happened.

The risk is particularly acute for practitioners entering domains for the first time inside AI-augmented environments. The senior developer who uses Claude brings decades of acquired competence that the tool amplifies. The junior developer who begins her career with Claude may develop genuine expertise in directing AI without developing the acquired competence of software engineering itself. The latter is not equivalent to the former. It is a different Discourse — the Discourse of AI-augmented practice — which is a real competence but not a replacement for the underlying domain mastery it depends on.

Origin

Gee adapted the acquisition/learning distinction from Stephen Krashen's work on second-language acquisition, where Krashen distinguished the unconscious, immersion-based process by which children acquire their first language from the conscious, rule-based process by which adults learn additional languages. Gee extended the distinction to all complex competences in Social Linguistics and Literacies (1990), arguing that the distinction applies not only to language but to any Discourse — including the professional Discourses through which practitioners become recognized members of their communities of practice.

Key Ideas

Two routes to competence. Acquisition through immersion produces tacit fluency; learning through instruction produces explicit rule-following.

Fluid vs. effortful. Acquired competence is automatic and robust; learned competence is conscious and conditional.

Identity integration. Acquisition produces competence that is part of who the practitioner is; learning produces competence the practitioner has.

AI extends learning. External scaffolding allows practitioners to apply rules without internalizing them.

Breakdown under stress. Learned competence fails when conditions diverge from training examples; acquired competence adapts.

Debates & Critiques

Whether AI can support rather than substitute for acquisition — functioning as scaffolding that accelerates the practitioner's passage toward internalized fluency rather than as a permanent external prop — is an open design question. The answer depends on whether AI tools are built to withdraw support as the practitioner develops competence (as good tutors do) or to maintain constant support indefinitely (as most tools currently do). The default market incentive favors the latter, which is why deliberate design choices — and institutional willingness to tolerate the short-term inefficiency of acquisition-supporting tools — will determine the outcome.