Research Article
The effortless trap: Productive struggle, AI, and the illusion of learning
Synthesis: Brcic & Frljic (2026) argue that the "allow or ban AI" framing is a false dichotomy; the relevant design question is placement. Used well, AI scales feedback, examples, practice, and individualized support; used poorly, it replaces the cognitive work learning requires and leaves an "illusion of learning" — a confident sense of mastery that collapses on the unaided task. The strongest causal evidence shows the outcome flips on design: an unguarded AI helper left high-school students ~17% worse on an unaided exam than peers with no tool, while the same model rebuilt to withhold answers erased the harm, and a well-engineered tutor roughly doubled learning. The authors give educators a graspable six-move frame (Prime, Probe, Point, Attach, Strengthen, Test) for placing the tool, with a one-line diagnostic: if letting AI in makes the task feel effortless, it is in the wrong place.
Key Findings
- Placement, not allow-or-ban, is the design question: AI belongs wherever it increases feedback, practice density, or realism without obscuring evidence that the student can think and perform unaided.
- Causal evidence flips on design: unguarded AI helper → ~17% worse on an unaided exam; same model withholding answers → harm erased; well-engineered tutor → roughly doubled learning.
- The six moves (Prime, Probe, Point, Attach, Strengthen, Test) model how one idea takes root, unifying productive failure/problem-before-instruction, worked-example, Socratic/cognitive-apprenticeship, deliberate-practice, and retrieval/self-explanation literatures.
- Protected moments: Probe (first hard attempt) and Test (final unaided check) are AI-out; Point/Attach/Strengthen allow guarded AI (hints, examples, practice); Prime is low-risk.
- Diagnostic: "If letting AI in makes the task feel effortless, it is in the wrong place" — but effort on the skill itself matters; AI should clear away the busywork that is not the skill (looking up, formatting, dead ends).
- Placement as governance: an AI-use policy becomes a per-skill design principle rather than a blanket prohibition list; the secured final check is the load-bearing point for grade and credential integrity.
What this means for practice
- Instructors. Place AI per skill rather than per course: keep it out of the first hard attempt (Probe) and the final unaided check (Test), and license it in between (Point, Attach, Strengthen) for hints, examples, feedback, and drill.
- Instructors. Judge each task with the frame's diagnostic — if letting AI in makes the task feel effortless, it is in the wrong place — while checking that the effort removed is busywork (looking up, formatting, dead ends) rather than the skill itself.
- Administrators. Replace allow-or-ban policy with a defensible per-skill placement rule, because the secured final check is what defends the credibility of a grade and of the credential behind it.
- Instructors. Start with one lesson and one skill before redesigning a course: hold the tool out of the first attempt and the final check, let it scaffold the middle, and reuse the six-move vocabulary across the courses students take in sequence.
Limitations
- The paper is a practical synthesis with no new empirical results: its argumentative weight rests on established cognitive and social science, while the AI-specific layer (2023–2026) is fast-moving, heterogeneous, and partly made of preprints.
- The frame covers one idea, on first encounter, in a single pass; durable retention through spacing and review is a layer above it, and the assessment architecture, AI-use contracts, syllabus templates, and per-course blueprints are explicitly out of scope.
- Several of its anchors are thin by the authors' own account: the access-timing study is a single lab with N = 105, the cognitive offloading EEG work it cites is small and contested and must be read alongside its published critique, and the tutor "doubling" comes from one elite crossover course rather than a general magnitude.
- The motivating master-teacher cases (Aristotle, Keller, Rátz, Szubartowski) illustrate what one-to-one mentoring can achieve but cannot establish causation — selection and survivorship effects dominate and the figures are non-peer-reviewed — and the model assumes the motivated, engaged student it cannot itself supply.
Citation
Brcic, M., & Frljic, S. (2026). The effortless trap: Productive struggle, AI, and the illusion of learning.