Answers are grounded in this paper and cite their sources.
Studies have examined the impact of Generative AI (GenAI) on students’ learning outcomes, revealing a wide range of effects on performance, cognitive load, and learning processes. However, little is known about whether these effects change with learner expertise and the conditions under which GenAI support is provided. This study examined expertise reversal in GenAI-supported programming education across four eight-week experiments involving Year 3 computer science students from research universities in Saudi Arabia. Drawing on cognitive load theory, the experiments investigated GenAI support, guidance level, task complexity, and access condition in relation to intrinsic cognitive load, extraneous cognitive load, mental effort, and independent programming performance. The findings showed that GenAI support produced greater reductions in extraneous load and stronger independent performance among lower-expertise learners, while higher-expertise learners retained smaller benefits. Detailed guidance produced a crossover expertise reversal, benefiting lower-expertise learners while increasing extraneous load and weakening performance among higher-expertise learners. Higher task complexity increased cognitive demands and reduced performance across expertise levels, with greater effects among lower-expertise learners. Personalised access also benefited lower-expertise learners more strongly without disadvantaging higher-expertise learners. Overall, the findings distinguish a crossover expertise reversal under detailed GenAI guidance from diminishing benefits as learner expertise increases under other support and access conditions. The effects of task complexity further demonstrate that expertise-dependent differences do not necessarily constitute a reversal in instructional effectiveness. We recommend that effective GenAI support requires careful calibration to learners’ existing knowledge and the cognitive demands of learning.
No takes yet. Share an insight, caveat, or question.
Yusuf et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: