Scaffolding vs Substitution

In the AI Story

The scaffolding concept comes from the Russian psychologist Lev Vygotsky, who argued that learning occurs in the zone between what a learner can do independently and what the learner can do with support. Below the zone, the learner is repeating what she already knows. Above it, she is overwhelmed. Within it, with appropriate scaffolding, she is stretched just enough to build new capacities. The essential feature of scaffolding is that it supports the learner's own cognitive construction without replacing it. The parent holding the bicycle seat is not riding the bicycle. The teacher providing the opening sentence is not writing the essay. The scaffold provides structure; the agency remains with the learner.

A child who uses AI to generate answers is receiving a substitute. The output has arrived — the completed essay, the solved problem, the explained concept — but the child has not undergone the cognitive construction that would have produced understanding. The building has appeared without the scaffold ever being erected, which means the building was not built by the child at all. It was delivered. A child who uses AI to generate better questions is receiving a scaffold. The tool helps her identify what she does not understand, formulate hypotheses she can investigate, articulate the gap in her knowledge that needs closing. The cognitive work remains hers.

The practical challenge is that the same tool and the same interaction can function as either, depending on the cognitive stance of the user. A student asking Claude to explain photosynthesis might be seeking raw material for her own model-building — or copying the explanation without engaging with the content. The difference lies not in the action but in what is happening inside the learner's mind: is she constructing, or receiving? The distinction cannot be enforced by monitoring every interaction; the developmental evidence is unambiguous that excessive control undermines the capacities parents most want to develop.

What can be done is to shape conditions that favor the scaffold use over the substitute use. Gopnik's research suggests several strategies: modeling curiosity (parents and teachers asking genuine questions); protecting unstructured time (allowing the boredom that default-mode processing requires); teaching evaluation rather than prohibition (examining AI outputs together, asking 'how would we know if this is right?'); and cultivating a relationship with difficulty (helping the learner persist through productive struggle rather than seeking instant rescue). Each strategy is a scaffold for the capacity to use AI as a scaffold — a meta-scaffold that the age of AI makes urgently necessary.

Origin

The concept of scaffolding was formalized by Jerome Bruner, David Wood, and Gail Ross in a 1976 paper building on Vygotsky's earlier work. Gopnik's application of the distinction to AI has been developed across talks, op-eds, and interviews since the early 2020s, with the clearest formulations appearing in her work for the Wall Street Journal and in her contributions to the 2025 Science paper on cultural technologies. The framework is directly operationalized in debates about AI in education.

Key Ideas

Same tool, different outcomes. Whether AI scaffolds or substitutes depends on the cognitive stance of the user, not the tool itself.

Substitution delivers outputs without capacity. The building appears; the scaffold was never erected; the learner did not build.

Scaffolding supports construction. The tool helps the learner do what she could not do alone in a way that develops the capacity to eventually do it alone.

Four protective strategies. Model curiosity; protect unstructured time; teach evaluation over prohibition; cultivate relationship with difficulty.

The meta-scaffold. The age of AI requires scaffolding the capacity to use AI as a scaffold — a second-order developmental task parents and teachers must now perform.

Debates & Critiques

Some educational theorists argue that the scaffold-substitute distinction is too binary, that in practice most AI interactions fall on a continuum, and that insisting on the distinction may produce excessive anxiety in parents and teachers whose children are using AI tools in mixed ways. Gopnik's defense is that the binary is not meant as a moral judgment on individual interactions but as an organizing question that parents and teachers should be able to ask at any moment. The question is actionable; a continuum is not.