Artificial intelligence in education is putting us through a pedagogical stress test

One of the recurring themes in current news is a fundamental pedagogical question: how to ensure that students remain the authors of their own learning. At a time when answers are always at our fingertips, it is becoming crucial whether students are merely users of the system in the learning process or whether they remain the agents of decision-making, judgment, and responsibility. Articles on this topic point out that a student’s active and co-responsible role does not happen on its own, but must be systematically built through tasks, questions, and learning situations in which the AI does not assume the role of author, but rather that of a conversation partner or support tool (for more, see the article The Key Podcast: Teaching Students Agency in the Age of AI). A similar emphasis can also be found in recent empirical research on the use of conversational systems to support learning, which shows that AI contributes most when it encourages further reflection rather than offering definitive solutions (more in the article Exploring a Conversational AI System in Supporting Children’s Literacy Learning at Home). We invite you to consider where, in your subject, the final answer is less important than the process through which the student arrives at it, and which part of that process should remain within the human domain?

In this context, AI very quickly reveals another sensitive issue in education: the question of rigor. Recent articles point out that AI does not undermine academic standards, but often reveals that these standards were already weak or superficial in practice. If assignments primarily test the form of writing or the ability to reproduce familiar structures, then generative models can easily produce work that is formally correct but substantively empty. This does not indicate that artificial intelligence is overpowered; rather, it suggests that assignments were often not designed to require higher levels of thinking from students (more in the article How AI Is Exploding Our Illusions of Rigor). In this context, the debate on the so-called “answer economy” is shifting toward the question of how to integrate judgment, interpretation, and justification into the learning process as central learning objectives (more in the article How Higher Ed Can Adjust to the AI-Answer Economy).

Such a shift, however, is not without consequences for students. Several articles highlight the so-called hidden costs of strategies that implement AI inconsistently or without clear pedagogical frameworks. Differing rules across courses, unclear expectations, and implicit prohibitions create an additional cognitive and emotional burden that often remains invisible but is very tangibly felt by students (see the article The Hidden Tax Students Are Paying for Your AI Strategy (or Lack Thereof)). This is also confirmed by research on the perceptions of higher education faculty, who often experience AI as a source of additional pressure rather than a relief, especially when there is a lack of systemic support and a shared understanding of objectives (more in the article Survey: Faculty Say AI Is Impactful - but Not In a Good Way). In this context, the question arises as to what extent existing forms of assessment measure higher-order cognitive processes and to what extent they primarily assess mastery of repeatable procedures. Assuming that artificial intelligence will take over routine tasks, it makes sense to consider which student achievements remain pedagogically and evaluatively relevant.

The debate on AI in education is therefore increasingly intertwined with issues of equity and accessibility. Articles on the new era of accessibility emphasize that digital and AI-supported environments are not neutral and that, without careful design, they can reproduce or even deepen existing inequalities (for more, see the article Higher Ed Prepares for New Era of Accessibility). This concern is further underscored by research on biases in language models, which warns that social inequalities can be imperceptibly embedded in the explanations and examples that AI offers to users (see the article Nature: AIs are biased toward some Indian castes — how can researchers fix this?). As AI becomes a partner in interpreting content, the question arises: who takes responsibility for evaluating its responses? Are we consciously creating space within our courses for students to learn how to critically assess the explanations provided by AI?

A special section of today’s newsletter features articles on the information environment in which students study today. Examples show that systems which confidently summarize and cite sources create an impression of credibility that is not necessarily well-founded. When AI cites content of questionable quality as a source, or when summaries are based on visibility rather than expertise, the distinction between citation and verification becomes a critical pedagogical issue (see the articles Latest ChatGPT model uses Grokipedia as source, Google AI Overviews cite YouTube more than any medical site). In a broader social context, this is complemented by warnings about agent systems and chatbots that can artificially create the appearance of consensus and thereby influence the formation of public opinion and debate (see the article Experts warn of threat to democracy from AI bot swarms). The example of the artificially created character “Amelie” further illustrates how quickly generated content can become a tool for spreading messages and, at the same time, a concrete example for reflecting on digital and media literacy (more in the article Meet ‘Amelia,’ the AI-generated schoolgirl). In an educational context, this once again raises the question of the difference between citing and verifying—do we make it clear enough to students that citing a source does not automatically mean it is reliable, and do we create opportunities in assignments where they must justify why the source they are citing is trustworthy?

The empirical studies included in this newsletter show that the impact of AI on learning is significantly more complex than is often assumed. Studies on chatbots in computer science education and on support systems for lab work confirm that AI can reduce frustration and improve task performance, but this does not necessarily imply a deeper understanding (see the articles Evaluating lab assistant chatbot on student learning and behaviors, Less stress, better scores, same learning). Similarly, research on predicting student performance using artificial intelligence raises questions regarding transparency, purpose, and the ethical use of data (see the article Predicting student performance: A comprehensive review of machine learning, deep learning, and explainable AI approaches).

In this context, teacher literacy is also becoming increasingly important. A review of research on AI literacy for educators emphasizes that the effective and responsible use of AI is not possible without a shared understanding of the technology’s fundamental concepts, limitations, and pedagogical implications (see the article Enhancing AI literacy for educators). Recent analyses of research on generative AI in higher education further demonstrate how rapidly the field is evolving and how unevenly distributed the evidence, experiences, and impacts are (see the article Generative AI in Higher Education: A Bibliometric Review).

Recent articles in the scientific journal *Nature* extend this discussion to the research and institutional levels. They point out that AI can simultaneously increase productivity and narrow the focus of attention, and that specialized model training can lead to unexpected systemic effects (for more, see the articles Nature: AI expands scientists’ impact but contracts focus, Nature: narrow training leads to broad misalignment). On the other hand, they also offer examples of collective intelligence, where human and machine knowledge complement each other in a way that keeps responsibility and judgment in human hands (more in the Nature article: “Collective intelligence and AI”).

In this newsletter, we wanted to emphasize that artificial intelligence in higher education is not merely a technical innovation, but rather a kind of “pedagogical stress test.” It does not answer the question of how to teach, but rather forces us to define more precisely what we want students to know, understand, and be able to evaluate. In this sense, AI is not a shortcut but, as we have previously pointed out, a mirror; and the question is no longer whether we will look into it, but what we will do with what we see there.

Authors: Sanja Jedrinović, Mateja Bevčič, Eva Kern Nanut, Eva Škraba, University of Ljubljana Center for the use of ICT in pedagogical process