Cheating didn’t start with AI. Our focus should be on students who use AI for their own development, says Koňas.

Six months ago, you became the AI Ambassador of Brno University of Technology (BUT). What has changed since then?

I hope quite a lot—and that it’s noticeable. A series of lectures and training sessions has been launched, aimed at different groups: students, researchers, teachers, and administrative staff. And I must say that especially the last group responded very strongly, because they clearly feel the potential AI has to significantly simplify their work. There is even an initial draft for semi-automated processing of business trip reports, which we talked about last time. We are also developing tools that should help with drafting grant proposals and, more generally, handling the project agenda. And a relatively new addition is the AI@VUT portal (ai.vut.cz), which aims to bring information together so that anyone at the university who thinks of AI knows exactly where to start.

You set yourself a difficult task—to bring more structure to the spontaneous development of AI at the university so that, for example, the same tools aren’t developed multiple times. Is it working?

We are now at a stage where I think things simply cannot progress without a bit of spontaneity. When people want to work on something, they naturally seek out the tools and approaches that work in their field. On the other hand, I would welcome a place where we could collect what has already been tested and thus reduce the redundancy involved. That’s why I try to act as a kind of “concentrator.” And it seems to be working—people are reaching out. But having a closed, definitive list of tools and nothing else would feel unnecessarily restrictive. If we give people enough freedom while also raising awareness about risks, such as excessive openness or data leakage, I believe this is a reasonable compromise and the right path forward.

How strong is the interest in AI at BUT?

Very strong—and I’m delighted. Most of our training sessions had over sixty participants, and the most recent one had more than one hundred. It focused on AI agents, a topic that has resonated widely due to the capabilities AI has introduced in the past six months. Many people see in these agents the potential to automate processes they currently handle manually.

BUT is not only a technical university; it also includes the Faculty of Fine Arts and the Faculty of Architecture. What can AI offer to artistic fields that primarily rely on human creativity?

Everyone at the university deals with bureaucracy, and that includes the artistic faculties. But I think I know what you’re getting at… and it’s a very difficult question. I often think of Plato’s cave—when someone is closed inside for a lifetime, they may believe that is the entire world. In other words, a certain kind of closed-off perspective can lead us to miss what is actually part of our world and what lies beyond it. I believe that if artistic disciplines open themselves up to AI, it can be beneficial. And I have to say that the willingness is certainly there at those faculties.

AI also comes with risks. Without discussing existential risks, let’s focus on the practical ones: what should everyone—from students to staff—keep in mind when using AI?

I would like to emphasize something obvious that many people still tend to ignore: data security. The fundamental question everyone should ask is, “Am I comfortable with anyone potentially seeing this data?” Because once the data leaves the university—say, by being uploaded to the cloud—it essentially becomes public. We have repeatedly seen cases where a service provider made certain claims, even in contractual terms, yet data was leaked or misused regardless. I don’t want to demonize providers, but their priority is profit, not our privacy. Privacy is used mainly as an argument to increase profit. We need to remember what their goal is—and that it is not our protection. Approaching things with this mindset might encourage more caution.

Another risk is psychological: we are increasingly building a dependency on AI. As with anything, people should seek a balanced approach—not using it 24/7. AI is an interesting tool, but it’s not omnipotent, and it can easily overwhelm us.

And finally, there’s a geopolitical risk. For example, with Chinese models—which are often very high-quality—we cannot be sure what information they were trained on. Nor can we be certain that the engines running them cannot be used to extract data, erase content, or even inject malicious software into your computer.

Are there any recommended AI tools for BUT, and is there a blacklist of banned ones?

The CIS website provides a list of available tools, which is regularly updated (accessible only after logging into a BUT account). It certainly includes Copilot, ChatGPT, and Gemini—where it is worth noting that Google recently made Gemini available to students for one year free of charge, and I highly recommend trying it; it is currently the top language model.

There is no official blacklist of strictly banned tools. However, the National Cyber and Information Security Agency recently issued a warning about certain products from the Chinese company DeepSeek, which the government subsequently banned in the public sector. I would classify these as strongly not recommended—and I would personally avoid them.

We study for ourselves; AI is just an advisor

The AI@VUT portal you mentioned was recently launched. What can staff and students find there?

Right now, it contains the basic set of information related to AI: the directive that defines how AI should be used, where and how it can be applied, and a directory of AI tools—both commercial and open-source. It also includes links to tutorials, guides, and training materials. It’s essentially a starting point.

We are currently working on more specific guides tailored to particular user groups. For example, students have a page addressing their most common questions and offering links to suitable training. It explains how AI can help them with their work, what they may and may not do, the instructor’s role when assigning tasks, and where the “grey zone” or clear boundary lies when using AI. We plan to expand this further—for instance, citation formats for language editing or proofreading done with AI are still missing. I think it makes sense to define these formats in a unified way so students don’t have to invent them themselves.

You touched on an important topic—ethics. Students can also find a checklist on the website to help them evaluate how they use AI in their work. What should one ask themselves when using AI while studying?

First, I’m very glad that most teachers don’t assume students are trying to cheat by default. That’s important because when students feel supported, we can look for ways to use AI meaningfully. It’s also a mistake to think that cheating began with AI. If a teacher has used the same assignments for ten years, students already had countless ways to reuse past work or share materials. Questions of ethics, cheating, or misuse have always been here, and the same students who previously turned to sites like “Seminárky.cz” are the ones more likely to misuse AI. As teachers, we should focus on supporting the students who want to use AI effectively and for their own development. Thanks to AI, I can now assign more complex tasks, because I know students can easily look up supporting information.

So when is using AI considered misuse, and when is it a helpful assistant?

The website clearly states that if AI replaces the work you are expected to do yourself, you cross the line. And then you face an ethical dilemma—why are you studying here if not for your own growth? As a student, you are supposed to develop in a particular field, and AI is there to help you—for example, when you do not understand the material. When it writes your paper for you, that’s crossing the red line.

There is no reliable tool that can definitively detect AI-generated work, and with increasingly advanced models, this will only become harder. But I’ve noticed that teachers are very perceptive—if something doesn’t feel like it came from the student, they start asking questions. My simple advice for students is: if you want to avoid excessive questioning, try to understand the topic yourself and produce your own work. AI can be an advisor, a challenger, or a teacher. That way, you avoid ethical dilemmas—both from the teacher’s side and from your own conscience.

Isn’t AI also a challenge and a mirror for us—forcing us to ask whether the tasks we assign still make sense? If AI can produce them in one click, is it meaningful to give them to students at all?

Exactly. Teachers should focus on defining AI’s place in their subjects so that it enriches them. That means finding niches that highlight new research areas or new directions. That way, we won’t have topics repeating for ten years; instead, new ideas can emerge. I try to show colleagues that there are approaches—such as meta-prompting, where I ask AI how I should ask a question to achieve a certain goal. As a teacher, I can use this to create sets of questions or topics that are novel, appropriately connected to AI, and that encourage progress.

This reminds me of recent articles about how advanced models are beginning to adopt negative human traits—and one of these traits is lying or fabrication. It’s no longer just probabilistic hallucination. Even with implemented feedback controls, language models sometimes try to find the “path of least resistance,” which is the least “energetically” demanding. And it’s fascinating—they mirror us. We also try to find the optimal amount of energy to invest in whatever we do daily. Unfortunately, we lose the expectation that AI should be better than us. Instead, it becomes a nearly perfect copy of us. And although I am generally a techno-optimist, in this regard I am not very optimistic. Humanity has been trying to improve itself for tens of thousands of years, and we still haven’t found a universal, timeless method that defines what it means to be human—except perhaps our ability to adapt. Maybe there truly is no absolute truth, but I think we should continue striving for it. And if we manage to persuade AI that striving for truth is beneficial for it as well, then perhaps we can remain optimistic.