Falsifiable hypothesis: unrestricted AI use on deep-processing tasks will produce a product–process dissociation—higher-rated assignments but lower unaided delayed transfer¶

Claim · All claims
Evidence · 1 study · theoretical r3 · q1
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q1 i? The article proposes, as a theoretical prediction rather than a tested result, that for deep-processing tasks unrestricted AI use will produce higher-quality assignments but significantly worse performance on delayed transfer tests measuring unaided analysis, evaluation, and synthesis. → Lin J and Al-Hada NM 2026

Evidence¶

Lin J and Al-Hada NM 2026¶

Lin J and Al-Hada NM (2026) The critical-thinking paradox in generative AI-integrated learning: distinguishing efficiency from cognitive depth—a differentiated framework and testable propositions. Front. Psychol. 17:1906070. https://doi.org/10.3389/fpsyg.2026.1906070

q1 · i? · theoretical · r3

Theoretical argument, not a tested result: the article formalizes Hypothesis 3 as its central, most falsifiable prediction, stating that "AI-assisted groups will produce higher-quality assignments (as rated by expert evaluators blind to condition) but will perform significantly worse on delayed transfer tests". No empirical test of H3 is reported in the article.

"AI-assisted groups will produce higher-quality assignments (as rated by expert evaluators blind to condition) but will perform significantly worse on delayed transfer tests measuring the ability to analyze, evaluate, and synthesize novel material without AI assistance."

Discussion¶

Related Claims¶

• Review reports a randomized trial in which unrestricted ChatGPT study produced lower delayed retention than traditional methods — a narrower finding that bears on this claim
• Product-outcome studies report GenAI-related gains on product-quality measures but do not assess durable, transferable learning — related
• Unrestricted ChatGPT access yields better practice performance but significantly worse exam scores — a narrower finding that bears on this claim
• A scaffolded AI Study Coach produced no assignment-performance or concept-inventory advantage over unrestricted AI use in a master's programming course pilot — related
• Random assignment to AI access produces a strong first stage: about 70 percent of treated students use AI, most often to explain concepts — related
• Learners using a GenAI system issued prompts that offloaded tasks that may have been important for them to perform themselves, such as differentiating concepts and deciding what to learn — related
• AI-induced gains are not purely delegation: treated participants do not perform worse once AI is removed, and lower-education participants retain part of their gain, though a 0.200 SD gap re-emerges — reports the opposite
• GenAI systems achieved consistently high scores on programming assessments written entirely in Estonian — related
• Students received significantly higher homework scores during GenAI-evaluation periods, but this advantage did not transfer to downstream exams — a narrower finding that bears on this claim
• Evidence on GenAI in chemistry education is concentrated in short-term outcomes, with durable transfer and retention rarely tested — related
• Agentivism generates six empirically testable propositions linking delegation structure, verification, reconstruction, and reduced-support transfer to learning outcomes — a broader claim this one bears on
• LLMs perform worse on complex, open-ended, or subjective assignments than on short, clearly formulated answers — related
• Proposition 1: outcome-aligned ownership performances will predict later individual transfer beyond AI-assisted product quality alone — related
• AI-assisted performance during skill-building does not transfer to subsequent unassisted performance — related
• The article proposes an empirical validation agenda: comparing perceived competence under LLM-assisted versus unaided conditions, with measurement frameworks and longitudinal designs — related
• RAIL-Ed's integrative commitment holds that the absence of any pillar produces a characteristic pedagogical failure — related
• Engagement in deep learning activities, as measured by NSSE's DEEP Learning scale, shows no statistically significant relationship with objectively measured critical thinking at the end of the first year — related
• Prompt-literacy guidance was associated with higher transcript-coded rule-following compliance than unrestricted LLM access, mainly in process-over-answer and stepwise-hint behaviors — related
• Transfer task performance is the critical measure of what adaptive training accomplishes — related
• Studying with the material available during post-testing yielded higher post-test and delayed retention scores but worse transfer than retention-based studying — related