Project-based learning and artificial intelligence (PBL-AI) in higher education

Generative AI reopens a familiar pedagogical challenge: integrating external tools without eroding authorship or the epistemic quality of learning. We propose and examine a PBL-AI model that treats AI as non-substitutive scaffolding, drawing on scaffolding theory, distributed cognition/extended mind, and cognitive load theory to delimit usage windows and fade support as expertise grows. The model combines explicit rubric-based criteria, a PBL sequence limiting AI to ideation, planning, and review, a traceability portfolio (prompt → AI output → verification → decision → changes), and a public defense without AI. Implemented in 2024–2025 at two Spanish universities – ULL (86 students) and US (198) – the study adopts a qualitative-interpretive approach, triangulating field notes, interviews, and documents (portfolios, rubrics, transcripts) and analyzing data through reflexive thematic analysis. Findings indicate clearer problem framing, more realistic planning, and stronger argumentative quality, alongside explicit authorship and auditable traceability. We discuss transferability and low organizational cost as enablers for institutional adoption.