Building on the concerns raised in Part 1, we now examine how the presence of AI in the classroom risks replacing genuine presence with polished performance. The subtle traces of real learning doubt, recovery, argument, and error are once again at risk of erasure.
The arrival of generative AI in education brings an unexpected gift: it places the fundamentals of teaching and learning back on the table. Suddenly, classrooms are forced to revisit the ancient questions what is learning, and what counts as understanding not as abstractions, but as urgent, practical challenges. This is not a loss. It is an opportunity. AI is not just a tool; it is a mirror that exposes what we have taken for granted and what we have left unexamined.
But the risk is not new. We have seen this before. When the first wave of digital devices and educational apps promised transformation, schools raced to adopt dashboards, adaptive platforms, and open repositories. With our attention drawn to the latest innovations, we drifted from our core. Disciplinary knowledge, pedagogical expertise, and didactic judgment were set aside. 'Everything is online anyway.' The teacher’s craft was quietly devalued, replaced by the false assurance of efficiency, and the profession lost sight of its own hard-won lessons.
Now, the cycle threatens to repeat. AI’s promise of seamless output and effortless personalization risks overshadowing the question of who is doing the work who is thinking, deciding, reflecting. In this environment, the subtle traces of authentic learning error, recovery, argument, and doubt are again endangered. Output becomes performance, and performance is mistaken for presence. As the recent article 'The Illusion of Thinking' illustrates, even the most advanced reasoning models can generate impressive traces of thought, but under true complexity their process collapses, leaving only the shell of thinking. This raises a crucial question for all of education: if nothing remains when support is withdrawn, was there ever real learning to begin with?
The danger is compounded by persistent educational myths. Tools are now being built, at unprecedented scale, on the long-discredited theory of learning styles and on a shallow reading of Bloom’s taxonomy treating it as a linear staircase of cognitive steps rather than as a flexible, recursive map of learning. Books such as 'How Learning Happens' (Kirschner & Hendrick) and 'Understanding How We Learn' (Weinstein & Sumeracki) have shown that effective learning is rooted in a deep understanding of cognitive science, not in fleeting trends or oversimplified models. Yet evidence-informed practice is still too often sidelined in favor of solutions that are pedagogically naive but technically impressive. We are in danger of repeating, with AI, the mistakes we made when we let the internet define what counted as knowledge.
Yet there is a path forward. The single most powerful way to ensure that AI strengthens rather than supplants education is to reinvest in deep disciplinary and pedagogical expertise. The question is not how much AI can do, but how well teachers and learners can see, diagnose, interpret, and intervene in the space between question and answer. When didactic knowledge is foregrounded, AI becomes an ally in the struggle for agency and ownership not an engine of further displacement.
This does not mean ignoring technology. On the contrary, investing in technological knowledge is now both essential and legally obligatory. The AI Act requires schools and educators to understand the tools they use, their limitations, and their consequences. True professionalism demands not only critical engagement with educational theory, but a deep and responsible grasp of the digital landscape in which we now operate.
The paradox is clear. The stronger the foundation of evidence-informed teaching, the more constructively AI can serve the act of learning. But when tools are designed on fragile or outdated learning theories, they threaten more harm than any technical limitation of AI itself. Here, responsibility is not only technical, but epistemic and ethical. The act of learning is not a question of output or compliance. It is the visible, risky, and often messy negotiation of agency and authorship. AI, if we are wise, will not erase these traces, but help us see them more clearly, and demand, once again, that we ask not only what was learned, but who did the learning.
-Hans Visser
*Part 2 was translated by using ChatGPT
