AI and a balance between comfort, judgement and responsibility

In connection with the opening question, it is particularly significant to note that AI systems are already capable of playing a significant role in their own development. The debate over what it means for AI to be increasingly involved in building the next generations of AI is no longer merely a technical one. It is a question of control, transparency and institutions that must understand the technology well enough not to leave it solely to the logic of speed.

The European framework for labelling synthetic content carries a similarly broad social significance. The new code for labelling and identifying content created by AI shows that trust in digital content will increasingly depend on traceability. This has direct implications for education, as students, teachers and researchers will need to be able to assess not only what the content claims, but also how it was created.

AI is changing us not just because it can create content, but because it is changing our habits. The question of whether we would trust someone who blindly trusts artificial intelligence is, in fact, a question about a new form of literacy. It is not enough to know how to use the tool; we must also be able to recognise when we must not leave judgement to it.

This discussion also includes a warning that AI must not rob us of our own reason. This is almost standard in education. The proper use of AI must not reduce mental effort to such an extent that the only result of learning is a polished product without any understanding of how we arrived at it.

Environmental and infrastructural aspects are also becoming part of the same story. The question of how much water and energy is required to generate a text using genAI reminds us that the digital is not immaterial. As universities integrate genAI into their day-to-day work, they must consider their environmental footprint alongside pedagogical and ethical issues.

Whether AI will take away jobs, and how it will affect cash, services and the day-to-day economy, is no longer a distant topic of the future, but part of the public debate on how work, payment and trust in systems are changing. For universities, this is a signal that AI literacy cannot be limited to the technical use of tools. It must become part of broader civic and professional literacy.

Concerns about access to the most powerful AI systems open up yet another layer. If access to advanced AI models is becoming a geopolitical issue, then European universities face a dual challenge: they must develop a critical approach to the use of existing tools, whilst at the same time reflecting on European research, linguistic and data autonomy.

In this context, it is worth noting that the internet is becoming a space where automated systems are generating ever-increasing amounts of traffic. If the digital space is filling up with synthetic content, then the role of education is changing and, as we mentioned in the introduction, it is no longer simply a matter of searching for information, but of assessing the source, purpose, credibility and consequences of using that content.

Healthcare remains one of the areas where the promises of AI seem most tangible. News about the use of AI in the operating theatre and about a vaccine developed with the help of AI shows that AI is not merely a tool for generating text and images. It is also a research and clinical partner. But this is precisely why it is all the more important that the training of future professionals continues to emphasise an understanding of AI methods, limitations and responsibilities.

The same applies to the integration of AI into everyday objects, such as wearable technologies and clothing. As AI moves away from a visible interface and becomes part of our everyday environment, making judgements will also become more difficult. Users will not always know clearly when they are communicating with a tool, when with an algorithmic interpretation, and when with a human decision.

In higher education, genAI is still most evident in the writing, assessment and understanding of students’ work. The question of what thesis projects still mean if they can be written by a language model does not necessarily spell the end of written assignments. It does, however, mean that a written assignment in itself is no longer a sufficiently reliable indication of the path a student has taken to reach an understanding.

This is why proposals that shift assessment from the final product to the defence, the process and reflection are particularly important. The idea that oral presentations can complement assignments that require text production is one of the more practical approaches. A student may use a tool, but must demonstrate that they understand the argument they have prepared, can explain their decisions and respond to the teacher’s questions.

A similar view is supported by the article on specifying the use of genUI, tailored to the individual task. A general statement such as “I used AI” is usually insufficient. It is different if a student notes that they used UI to plan the structure, check the code, improve the language or create content. According to various authors, a more precise disclosure of AI usage should therefore become part of AI literacy.

In this context, we would like to reiterate a warning regarding detectors. If UI detection tools are inconsistent and require a rethinking of assessment, then universities cannot build integrity on the technical ‘hunting down’ of students. A more promising approach is to consider reforming assessment in a way that makes the thought process more visible and transparent.

Among the good practices highlighted over the past two weeks, one approach stands out: rather than banning AI for students, it helps them understand how to use it. A practical framework for students’ use of AI is important because it moves students out of a grey area and into a space of clear expectations – when its use is permitted, when it must be cited, and when its use would replace an essential learning objective.

Research into guided use of genUI in learning further demonstrates that the tool alone is not enough. The effect becomes apparent when the interaction is structured – when the student is guided to explain, verify, compare and draw conclusions about the answer generated by the AI.

The discussion regarding how an ‘all-or-nothing’ approach to AI can restrict scope for innovation is also highly relevant for universities. Complete bans often do not work, whilst complete unrestricted use obscures the actual learning objectives. Between these two extremes, there is scope for subject-specific rules.

In academic work, we also found the shift in literature searching particularly interesting. The report on the workshop held on generative AI and academic searching points out that search engines are no longer merely lists of sources. They are increasingly summarising, classifying and interpreting content. Consequently, activities that teach students to check which sources are included, what has been omitted from the AI results, and how reliable the synthesis produced by AI tools is, are once again proving to be good practice.

An increasing number of news stories share a common thread: the issue of cognitive offloading. If genAI takes over the task of drafting text, summarising and structuring an argument, the student may save time, but may lose precisely the effort that builds understanding. The discussion on writing, cognitive offloading and genUI is therefore not a nostalgic longing for more and harder work, but a fundamental question of which thought processes we wish to preserve in tasks where students can use genAI to assist them.

A similar emphasis is placed in the article on how the student environment has changed with genAI. Student life is changing not only in tasks where students use genAI, but also in learning, communication, seeking help and the sense of what independent work means. Universities must therefore consider pedagogical changes to complement existing regulations on the use of AI.

We would also like to add a warning to this chapter that students are still often the weak link in cyber security in higher education. With the rise of genAI, the risk of fake content, automated scams, fraudulent logins and data misuse is also increasing. Digital literacy must therefore also encompass literacy in this area.

The issue of stricter sanctions for research misconduct has become more pressing in the age of AI. Synthetic texts, fabricated references, automated writing and pressure to publish may increase the volume of seemingly academic texts, but not necessarily the quality of knowledge. This is also confirmed by the debate on the taxonomy of harm in the age of AI, which argues that harm is not merely a matter of incorrect answers, but also the loss of trust in the processes through which knowledge is generated.

Therefore, perhaps the most important message of today’s newsletter is this: genAI does not reduce the need for education, but quite the opposite. More clear learning objectives, more explanation, more assessment, more responsible task design and more discussion about what knowledge means.

The latest news portrays AI as a technology that simultaneously promises to accelerate progress and requires us to apply the brakes. It can assist with research, medicine, accessibility, writing and learning. However, it can also mask a lack of understanding, increase dependence on tools, obscure the origin of content and undermine trust in academic processes.

For higher education, this is yet another incentive to work with greater rigour. Universities will need to help students use AI in a way that enables them to think better, not less. This means less blind trust, less reliance on unreliable detectors, and more tasks in which process, judgement, responsibility and understanding are evident.

In this issue of the AI newsletter, we’d also like to invite you to get involved. If you’ve tried out an interesting way of using UI in education – whether at your faculty, in a course or as part of your professional work – you can share your example with us via the form for collecting examples of good practice. We look forward to reviewing the submissions and including selected contributions in future issues of the AI Newsletter, so that together we can build an overview of useful and well-considered approaches within the university sector.

If you have missed any issues of the AI newsletter, you can always view them on the page containing all AI newsletters.

Authors: Sanja Jedrinović Čufer, Mateja Bevčič, Eva Kern Nanut, Eva Škraba, Maja Kosmač, University of Ljubljana Centre for the Use of ICT in the Educational Process