Context. Since late 2022, the rapid uptake of generative AI (GenAI) tools such as GitHub Copilot and ChatGPT has outpaced computing educators' evidence base for responding to it. Objective. This systematic literature review synthesises empirical evidence on four interconnected dimensions of GenAI in computing and programming education: effects on learning outcomes, student over-reliance, assessment integrity, and pedagogical adaptation. Method. Following Kitchenham and Charters (2007) guidelines and Wohlin's (2014) snowballing from seven curated seed papers, screening yielded 72 primary studies (2022–2026) from 33 venues, each was quality-assessed on a 10-item checklist (mean 8.32/10). Thematic synthesis (Cruzes and Dyba, 2011) produced 14 themes. Results. Short-term efficiency gains are robustly replicated (36 studies) but are offset by a comprehension–completion trade-off: AI-assisted completion does not transfer to independent skill (21 studies) and is moderated by prior knowledge (6 studies). Passive over-reliance is widely documented (21 studies), with guardrail tools, self-regulated-learning scaffolding, and scaffold withdrawal as promising interventions (20 studies). Most standard assessments are AI-completable (13 studies), undisclosed AI use is widespread (45–65%; 13 studies), and AI-resistant assessment design is the most densely evidenced theme (26 studies). Conclusions. Responsible integration requires coordinated action across all four dimensions. We derive the VIE Framework (Verification, Implementation, and Equity): three interdependent design requirements, and treating any one in isolation underperforms. Seven priority research gaps are identified, notably the absence of longitudinal skill-development and cross-national equity studies.
- Inicie sesión para enviar comentarios