The cognitive offloading trade-off in generative ai-assisted mathematics learning: a model of optimal reliance
Keywords:
generative artificial intelligence, cognitive offloading, desirable difficulties, retention, optimal reliance, higher mathematics, didactic modelling, assessment validity, effect sizeAbstract
Generative AI raises measured performance while a student solves problems and may simultaneously reduce what is retained, because the cognitive work that produces durable competence is the work being offloaded. The policy question is therefore not whether to permit AI but how much reliance to permit. This paper models reliance as a continuous control variable entering a differential equation of competence formation through two opposing channels: a linear gain in engagement and coverage, and a convex penalty on retention. The model has an interior optimum at a* ≈ 0.38, raising unaided competence from 56.8 % to 61.1 %; reliance beyond a ≈ 0.70 is worse than using no AI at all, and unrestricted reliance lowers it by 17.7 %. Calibrated independently at two points to a published field experiment, the model also quantifies an assessment illusion: at full reliance, coursework scores exceed unaided competence by 37 percentage points
References
1. Decree of the President of the Republic of Uzbekistan No. PF-5847 of 8 October 2019 “On approval of the Concept for the Development of the Higher Education System until 2030”. – Tashkent: Lex.uz, 2019.
2. Bastani H., Bastani O., Sungu A., Ge H., Kabakcı Ö., Mariman R. Generative AI without guardrails can harm learning: Evidence from high school mathematics // Proceedings of the National Academy of Sciences. – 2025. – Vol. 122, No. 26. – e2422633122.
3. Kestin G., Miller K., Klales A., Milbourne T., Ponti G. AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting // Scientific Reports. – 2025. – Vol. 15. – Art. 97652.
4. Risko E.F., Gilbert S.J. Cognitive offloading // Trends in Cognitive Sciences. – 2016. – Vol. 20, No. 9. – P. 676–688.
5. Sparrow B., Liu J., Wegner D.M. Google effects on memory: Cognitive consequences of having information at our fingertips // Science. – 2011. – Vol. 333, No. 6043. – P. 776–778.
6. Roediger H.L., Karpicke J.D. Test-enhanced learning: Taking memory tests improves long-term retention // Psychological Science. – 2006. – Vol. 17, No. 3. – P. 249–255.
7. Bjork E.L., Bjork R.A. Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning // Psychology and the Real World. – New York: Worth Publishers, 2011. – P. 56–64.
8. Sweller J., Ayres P., Kalyuga S. Cognitive Load Theory. – New York: Springer, 2011. – 274 p.
9. Murre J.M.J., Dros J. Replication and analysis of Ebbinghaus’ forgetting curve // PLoS ONE. – 2015. – Vol. 10, No. 7. – e0120644.
10. Cohen J. Statistical Power Analysis for the Behavioral Sciences. 2nd ed. – Hillsdale: Lawrence Erlbaum Associates, 1988. – 567 p.


