Following our recent look at the "Safety Gap" and the dangers of using AI tools that simply deliver immediate answers, it is worth asking what the alternative actually looks like in practice. The theoretical argument for "productive struggle" is compelling, but where are the examples of technology being used to genuinely support cognitive development rather than bypass it? A recent report from the UNESCO Courier offers a striking, concrete example from an unexpected context—and it holds profound lessons for vocational education and training in Europe.

In the mountainous Guizhou province of southwestern China, a project called Hongyan (meaning "wild goose") is being deployed in rural schools. As the UNESCO report details, the developers of this system deliberately chose not to use standard commercial large language models optimized for speed and instant answers [1]. Instead, they implemented a Socratic interaction logic designed specifically to protect the student's cognitive development. The system is explicitly programmed to "refuse the answer trap." When a student encounters a mathematical impasse, the AI tool does not provide the solution. Rather, it initiates a diagnostic dialogue, asking questions like "At which logical juncture did the path become unclear?"

This approach transforms the AI from an Oracle into a patient mentor. For the students in Guizhou, it provides an environment where they can practice without fear of judgment, allowing them to deconstruct their own thought processes. Crucially, by offloading the mechanical tasks of grading and foundational knowledge dissemination to the AI, the human teachers are liberated to reclaim their true vocation: providing emotional support, moral guidance, and noticing the subtle shifts in a student's confidence that an algorithm will always miss.

While this example comes from a rural primary school context, the pedagogical philosophy underpinning it is directly applicable to vocational education and training. In fact, we are beginning to see this exact approach—often termed "cognitive apprenticeship"—being integrated into advanced skills training. For instance, recent research from the National University of Singapore demonstrates the effectiveness of an "AI Learning Buddy" designed for undergraduate nursing students [2]. Much like the Hongyan project, this nursing tool was deliberately structured to simulate Socratic questioning.

Rather than allowing trainee nurses to rely on search engines for quick answers—a practice that weakens their relationship with authentic clinical environments—the AI Learning Buddy forces them to engage with established clinical reasoning cycles. It prompts them to reflect, analyse, and justify their clinical decisions. The preliminary findings from this study show that students primarily use the tool not to cheat on assignments, but to review complex clinical procedures and prepare for patient education during their ward-based placements [2]. It functions as a bridge between classroom theory and ward-based practice, mediating meaningful interactions rather than merely supplying decontextualised answers.

This is the model of AI integration that European VET systems should be pursuing. Whether we are training nurses, electricians, or advanced manufacturing technicians, the goal is not to make the task easier, but to make the learning deeper. When an apprentice encounters a complex fault in a piece of machinery, an AI assistant should not immediately highlight the broken component. It should ask the apprentice to articulate their diagnostic strategy, prompting them to eliminate possibilities logically, just as a master craftsperson would guide an apprentice on the shop floor.

The true innovation in educational AI is not the underlying code, but the ethical and pedagogical framework in which it is deployed. As the founders of the Chinese project noted, treating the student as someone capable of self-improvement, rather than a mere consumer of information, is the core requirement. By embracing Socratic AI tools that demand productive struggle, VET educators can ensure that technology amplifies, rather than replaces, the profound human endeavor of learning a trade.

References

[1] Zheng, W., An intelligent solution to inspire young minds in rural China, The UNESCO Courier, April 2026. https://courier.unesco.org/en/articles/intelligent-solution-inspire-young-minds-rural-china

[2] Siah, C. J. R., et al., AI Learning Buddy To Elevate Critical Thinking And Clinical Reasoning To Enhance Nursing Competence, Higher Education Conference in Singapore (HECS) 2025, National University of Singapore. https://ctlt.nus.edu.sg/hecs-2025/abstracts/hecs2025-lt-rosalind-s/

About the Image

A network diagram of diverse emojis and symbols to represent structured databases. The artwork reflects how AI processes vast information and emotional signals based on definable and relational attributes of the data. Machine learning algorithms can analyse structured data to extract patterns and classify information based on inferences from the data. Created in Inkscape, it explores the complexity of knowledge representation. The emojis used are from OpenMoji, which is the first open source and independent emoji system to date.