Artificial intelligence in vocational education and training: A systematic review of educational purposes, theoretical conceptualizations, and empirical effectiveness

Texto completo

Background Artificial intelligence (AI) increasingly shapes vocational education and training (VET). Despite its transformative potential, a systematic overview of the design characteristics and empirical effectiveness of AI-based interventions in vocational contexts remains lacking. Aim This review synthesizes empirical research on AI-supported interventions in VET, focusing on educational purposes, theoretical conceptualizations of human-AI interaction, methodological designs, and evidence of learning outcomes. Method Following PRISMA guidelines, 26 empirical studies published between 2015 and 2026 were identified through ERIC, Web of Science, and Elicit, and analyzed using a theory-informed coding scheme. Findings Intelligent Extended Reality shows consistent positive effects on procedural competence, practical skills, and learner motivation; Intelligent Tutoring Systems foster declarative and procedural knowledge; AI chatbots show promising effects on self-regulation and task performance. However, the evidence base is methodologically constrained: only five randomized experimental studies were identified. Across applications, AI is predominantly implemented through behaviorist or cognitively oriented instructional designs that emphasize drill-and-practice and adaptive feedback. In contrast, approaches fostering learner agency, critical reflection, and autonomous decision-making remain underrepresented. Conclusion Current research largely reflects a generalized „success narrative” surrounding AI in VET. Future studies should investigate failure cases, contextual moderators, and boundary conditions more systematically to develop a more differentiated understanding of the effectiveness of AI interventions. To realize the transformative potential of AI in VET, research and practice must move beyond replicating human instruction—avoiding the Turing Trap—and instead design learning environments that augment human judgment, strengthen learner agency, and support teachers in empathetic and holistic guidance.