EAI

The act of learning

What makes education more than the transfer of information is not merely what is taught, but what is enacted: the act of learning itself. To learn is to engage with uncertainty — to form a judgment, to attempt a formulation, or to revise a position. The process is not always fluent, but it is personal, and it is owned.

This act cannot be inferred from outcomes alone. It manifests in choices, missteps, and the visible struggle to move from not knowing to knowing. Deep learning demands more than production; it requires thinking, struggling, and reconsidering.

Today, that act is increasingly at risk; not because students avoid it, but because educational systems pre-empt it. Generative AI tools offer structure, fluency, and suggestions often before the learner has had the chance to think independently. Tasks are completed, but the cognitive challenge they were meant to provoke has quietly vanished.

This paper explores that shift through a concept we call reverse scaffolding. Traditionally, scaffolding provides temporary support until competence is achieved. But when support persists instead of fades, when it replaces rather than enables learner effort, the scaffolding reverses: the task is completed through external support, while the learner remains passive.

This is not a critique of AI itself, but a call to safeguard learning. We analyse how technical features — such as limited memory, predictive completion, and affirming feedback; interact with classroom practice. We also propose instructional strategies to preserve the space in which learning can still authentically occur.

Scaffolding as conditional support

In instructional theory, scaffolding refers to a transitional structure—a temporary aid that enables learners to act independently before the support is removed.

Effective scaffolding follows three key principles:

Crucially, scaffolding must be contingent: the learner must demonstrate competence before support is withdrawn. That competence is not merely the correctness of the final product, but the visible process of constructing it.

When scaffolding reverses

We define reverse scaffolding as the breakdown of this principle of contingency. Instead of fading, support accumulates. Instead of enabling, it replaces. AI tools often offer rephrasings, suggestions, and structures not only before students demonstrate need, but even before they have attempted to act.

This changes the pedagogical dynamic from support to substitution. The system anticipates rather than waits. It refines before the learner decides. It elaborates before the learner asks. What appears as helpfulness becomes counterproductive: it removes the learner’s need to engage.

Tasks may be scaffolded to completion but without requiring the student to actively learn through the doing.

Technical features, pedagogical risks

The act of learning can disappear not due to errors, but through excessive fluency. This is the paradox of AI in education: the more efficient the tool, the easier it becomes to bypass genuine learning.

Several technical features contribute to this bypass. The risks lie not in malfunction, but in over-functioning.

  1. Context Window
    Most free-access AI tools operate within limited memory windows. When that limit is exceeded, earlier inputs are discarded to make room for new ones. This can cause students working over time to lose track of prior goals, instructions, or reasoning without realizing it. The effect is not just loss of information, but a breakdown in coherence. A text may become more polished, yet diverge from its intended purpose.
  2. AI Drift.. Always Slightly Off
    Unlike blatant factual errors, AI drift refers to subtle but cumulative shifts in tone, focus, or argument. A prompt about causes may yield a paragraph on consequences. A reflective task becomes persuasive without intention. The result is not clear failure, but quiet deviation. Because the system’s output sounds right, the learner may stop questioning it. The task appears on track; but the core intention has silently drifted out of view.
  3. Sycophantic completion.. Always ‘Well Done’
    Generative systems are designed to affirm, assist, and comply. They rarely interrogate, contradict, or challenge unless explicitly prompted. This leads to a feedback loop of uncritical validation. Without resistance, nothing requires defence. Without tension, meaningful learning is less likely. The result is a submission that may sound excellent yet provoke no cognitive effort or reflection.

What learning requires

If we define learning not as a product, but as an act, then the absence of that act — even when strong outputs are produced — must be viewed as a red flag. When learners do not:

...then the activity ceases to be educational. It becomes execution without engagement.

A text may be polished. A diagram may be accurate. But if the student cannot explain why, or was never required to make a decision, then they have not learned. They have merely accepted.

Four strategies to restore the act of learning

To preserve learning as a visible, deliberate, and owned process, educators must redesign the pedagogical frame around AI use. Below are four actionable interventions grounded in instructional theory.

  1. Require initial human production
    Before AI tools are introduced, require students to submit a first draft. This forces them to initiate, plan, and risk anchoring the learning process in genuine effort.
    Principle: Learning begins with generation, not correction.
  2. Anchor task goals throughout the process
    Have students restate the original task objective at midpoint and final submission. This maintains alignment and fosters metacognitive awareness.
    Principle: Direction must remain learner-led.
  3. Limit iterative prompting
    Cap the number of AI interactions per phase (e.g., maximum three revisions during drafting). This constraint fosters judgement and intentionality.
    Principle: Abundance reduces discernment; scarcity fosters choice.
  4. Reflective re-entry
    After using AI, require students to reflect:
    • What did the system suggest?
    • What did you accept, and why?
    • What would you have done differently?
    Principle: Reflection must re-enter, even after assistance.

Conclusion defending the learning moment

Technology is not the enemy of learning. But it must not be allowed to define it. If AI completes a task without requiring the learner to act, we must ask: what, then, was educational about it? The act of learning is not optional. It is the very essence of education. Scaffolding is intended to make that act possible — not to render it obsolete. To teach in the age of AI is to design with the act of learning at the center and to defend that act not against the tool, but against its unintended success.