Cognitive Dependency in AI-Assisted Programming: Correlation Between GitHub Copilot Usage and Syntactic Memory Degradation in Engineering Students

Benites-Rodriguez J.; Benites-Angulo L.; Benites-Angulo C.; Tabacchi-Murillo J.; Rodriguez-Terrones J.; Amaro-Guzman C.; Torres-Quiroz A. · Proceedings of the LACCEI international Multi-conference for Engineering, Education and Technology (2026) · DOI: 10.18687/laccei2026.1.1.2421

Resumen

This longitudinal study examines the cognitive impact of GitHub Copilot on 1,460 engineering students. Using a mixed-methods approach combining cognitive load theory assessments, syntactic retention tests, and development speed metrics, we document a significant inverse relationship between AI assistant dependency and long-term syntactic memory consolidation (r = -0.67, p < 0.001). While intensive Copilot users demonstrated 34.2% faster task completion times initially, they exhibited 41.8% lower syntactic recall in delayed assessments (Week 16) and 53.6% reduced performance in unassisted programming conditions. Analysis revealed that intensive AI use correlates with decreased Germanic cognitive load (r = -0.54, p < 0.001), suggesting reduced deep processing essential for skill acquisition. Domain-specific analysis showed a pronounced deterioration in advanced constructs: object-oriented programming (-42.1%), functional programming (- 38.7%), and complex data structures (-37.3%). Structural equation modeling confirmed mediation through Germanic cognitive load (indirect effect = -0.33, 95% CI [-0.39, -0.27]), explaining 49% of the total relationship. These findings reveal a critical productivity-learning paradox in AI-assisted programming education, with implications for curriculum design and pedagogical practice in computer science education.

Instituciones: Universidad Nacional de Trujillo; Universidad Privada Antenor Orrego; Universidad Nacional del Callao