In many fields, we are limited by human cognitive possibilities. Artificial intelligence can help us process the vast amount of information available to mankind. This is what Petr Koňas, an expert in numerical simulations and AI, who has recently found himself in the role of ambassador for AI at BUT, believes in. Thanks to his view, he is a mild techno-optimist who hopes that the use of artificial intelligence will primarily benefit humanity, whether in the form of new revolutionary discoveries in science or by making it easier to fill out a travel expense report.

Introductory note: The interview goes far beyond the usual length of texts on the FSI website, because the topic is so broad and the answers so interesting that it would be a shame to shorten it forcibly. Please note in advance that this is a longer text. If you are mainly interested in the role of an AI ambassador at BUT, skip to the last third.

When I was preparing for our interview, I consulted the Perplexity research tool about you. What is your favourite tool in your daily work that makes your life and work easier?

There are a number of tools, depending on what I'm currently focusing on. Perplexity is also one of them. Then there is LibreChat, which is a kind of concentrator that brings together most commercial and open-source language models and offers the ability to switch from a specialised model to a more general one according to what you need at the moment. Within LibreChat, there is also a very useful option for so-called agents, which are tools that allow you to deal with narrowly specialised areas: for example, when I am searching for a literature review, I call the appropriate agent and my job is very simplified then. Recently, I have been using the RAGFlow tool for economic and administrative tasks within the Rector's Office, and we are now working with colleagues from the Rector's Office on the possibility of using AI in the field of accreditation.

And then it's Python in general, because many tools are available in a very raw state, and if I really want to use their high-end capabilities, I have no choice but to modify them. In this, Python plays a truly irreplaceable role.

Most of us have been interested in AI since we first clicked on ChatGPT, which is just a few years back at most. But you have been working in the field of artificial intelligence for 30 years. What kind of AI was it back then, and what development had it come through before it made it into the common vocabulary of today's population?

It's a huge change. The first moments with AI were relatively unremarkable for me. AI was one of the many statistical methods that were used for analysis, approximation, various classifications and so on. For example, it helped me a lot 30 years ago in image analysis, because it was able to perform image segmentation much more efficiently, for example. However, at that time, we encountered the fact that there was a lack of sufficient computing power. Each analysis, even though it shifted the quality of the evaluation, took an extremely long time, and you thought twice about letting the machine work on it for a month or even longer. But as hardware capabilities have changed and computing power has grown, so has the influence of AI, whether in machine learning, then deep learning, or current language models.

In particular, the use of language models in conjunction with machine learning and all the other possibilities that are offered en bloc today is absolutely irreplaceable and will play a huge role in the future. Because not only for us scientists, but basically for all of humanity, it offers a unique and fantastic opportunity to lead you to relevant data based on relatively vague information and find a solution that we have either not been able to find until now, or it took so long that we gave up on it. I'm very glad that I've lived to see the current form of AI and that there are even more unimaginable changes in this area.

I see that you are a techno-optimist, so I have to ask whether you think AI will bring us new and truly groundbreaking discoveries in fields where we have been waiting in vain for years, such as astrophysics or particle physics. Do you see this as one of the roles of AI?

That's exactly right. Let's realise that in many fields we are already encountering human possibilities, not only cognitive, but we are also encountering how to process the amount of information that occurs. Just think of our university – how many articles and knowledge BUT generates every year! There is no person who would be able to process the data in such a way as to evaluate all this knowledge. It's always just a fragment. Basically, we are not able to create synergistic knowledge that would really move us further, not to mention the consolidation of knowledge across different disciplines. I'm afraid that the human brain really doesn't have what it takes anymore, and we need to change it; we need tools to help us with this. AI has what it takes, it can go not only into great depth, but also across disciplines and discover connections that humans have no chance of noticing.

Recently, Tomáš Mikolov made a rather unfortunate statement about the "autism" of some scientists. But we have to admit that as researchers, we really get into roles where the rest of the population has no chance to understand not only the meaning of what we are saying, but also the application of our results. In this respect, AI can again play an irreplaceable role, as it can mediate our knowledge to the public.

I can confirm this because when I go to do an interview with, for example, colleagues in physics, I use AI to help me understand the scientific texts we are talking about. And AI really knows how to "translate" the texts in a popularizing way.

You are right, but it is also its great danger, that is, that we do not remain only "on the surface". We have to deal with how to move a person further in the context of knowledge. And this is a great opportunity for us as an educational institution: how can we move our students so that, even with the limitations that they have and that each of us has, they can master the subject in a certain time and quality. AI can help us here as well, because it has the capacity to attend to each student individually, which a live teacher often does not have.

Human imperfection

I don't want to dwell too much on the concerns that AI raises here, because there are a lot of them. But I'll ask one question: What do you think will remain with us as humans if we hand over many of our tasks and skills to AI?

That's a very difficult question. I think that a human being, like the rest of nature, is subject to evolution. And now we are hitting a number of our limits just because of how huge a leap humanity has made in the last hundred years. The shift is so extreme that our physiological conditions limit us terribly, they pull us down. And I ask myself how we can overcome this handicap, and I think that AI is again a natural step in this to move forward.

But you're right, many times I wonder if the development is not too fast and if we are ready for it. And it can be seen from the reactions of the general population that we are often not ready. We don't realise that what AI returns to us is basically an image of ourselves. And what we consider to be mistakes are the mistakes of humanity. We should not only work on adapting AI, but also look for the root causes of our problems. If we don't work on it, there is a risk that we will create a fundamentally more advanced form of our imperfection, and this is definitely not a positive outlook. But as a scientist, I am optimistic and believe that thanks to AI, we will be able to uncover the contradictions that we have as humanity and that we have so far overlooked. Perhaps it will allow for a kind of catharsis, an awareness of what humanity is and what defines us as a person.

Since you mentioned the rapid development, I will also ask about the concerns of some experts that, with the growing amount of generic content, AI may start to degenerate. In other words, it can get harmed when it learns from its own outputs. At the same time, it is clear that humanity's dependence on this technology will grow...

I understand the concerns of all data analysts who point this out. On the other hand, I would add that a number of scientific articles also bear certain features of copying and redundancy. And yet we dealt with it and used it as training data, as a basic standard. And we have no problem with that. Here I perceive a kind of internal contradiction, but, of course, the more of this redundant data there is, the less the AI will develop. It is a natural development, where quantity and quality alternate in different phases. And even within neural networks, we can see that in some places we need a large amount of data, and in others, a breakthrough discovery that will change the way it is processed or interpreted. That will move us forward.

At the moment, we are talking "only" about AI, i.e. artificial intelligence, not about so-called AGI (general artificial intelligence, which aims to replicate the cognitive abilities of humans, editor's note). When do you estimate that it will be developed? Or will we continue to use partial AI tools for specific purposes?

Very soon after the language models were released, it became clear that their limitation lies precisely in the way they see the world. In the case of textual language models, their modality is textual input. Then came speech processing, then image processing, and the number of modalities that describe the world around us is constantly increasing. The more it grows and the more they connect with each other, the more new relationships and a new kind of knowledge are revealed. And the closer we are to some kind of universal AI that can represent us in terms of perceiving the world as we perceive it. If this assumption is met, if we also manage to create something like a common motivation, I believe that AGI will be created in the relatively foreseeable future. However, by the foreseeable future, I mean 20 years or more.

Everyone deals with travel expense reports

Let’s leave the big questions and get our feet back on the ground: What does AI mean to you personally? And how do you find a relationship with it?

For me, it's a huge help because it simplifies my work. What used to take months, then weeks, is now minutes in which I am able, for example, to perform the aforementioned literary search with a single click. Thanks to this, I have time for things that I have been putting off for a long time. And I think that this is a great challenge for the university and for research in general – by eliminating bureaucracy and other activities that are unnecessary in relation to creative activity, we can then really focus on the creative activity itself, which I consider to be the main feature of our humanity.

Let's now look at your position as AI Coordinator at BUT. How did you actually get to it – of course, with regard to your expertise?

It was just sheer luck. (laughs) I was helping my colleagues at the institute, then it expanded into the faculty, and suddenly, Vice-Rector Vítězslav Máša contacted me that he had heard about me and was currently addressing the issue of AI strategy and development within the university. He has a very interesting, reasonable view on this. I wish everything we talked about could be implemented. Anyway, we made a deal and I started.

What should be your role?

The main idea is to ensure coordination, because everyone today has their own view of AI, using their own tools and procedures. Unfortunately, this leads to a lot of things being duplicated, and we waste time on what someone has already solved a long time ago. My role will be to unify information, create a knowledge base on AI, and be a contact person. We should accelerate developments in this area, because, let's not kid ourselves, Czech universities are a bit behind the West in this. Primarily, I understand my role as meeting people, creating teams focused on specific areas of AI, and helping them in the transition to AI. So that artificial intelligence is really a helper and not more of a problem it has to deal with.

Is the creation of a BUT AI Assistant being considered? For example, I ask how to fill in a travel expense report and I get advice?

When my colleagues and I started talking about AI, I thought that the main problem they would see would be the use of AI in research. But I couldn't be more wrong! The biggest problem that worries everyone is administration. And almost everyone mentioned the travel expenses. These are really things that are holding us back, and it would be good if they could be automated in some way. So yes, it's one of the things that plays a major role. But I'm not entirely sure if it has a simple and quick solution, because when you start reading through all the guidelines associated with it, you have a problem grasping it.

But you are right, and the companies that approach me most often ask about how to process and make available a large amount of information and corporate know-how so that it does not disappear with the departure of employees. AI is ideal for this because it can collect information, sort it, and easily guide you to your goal when asked.

What can BUT employees contact you with at the moment?

Definitely with the requests for training, I'm already doing it, and I'll continue to do so. But in general, this is how it is: if you are dealing with AI, get in touch. It doesn't matter if the problem is small or big… sometimes even a small problem can become a very interesting thing. I will be very happy to help anyone as much as I can.

At the same time, I don't claim that I have an answer to all questions, and I certainly don't know about everything that's going on in AI, although I try to be in on it all the time. So, I will be very happy to share my experience with other people.

You seem enthusiastic – what do you actually enjoy about AI?

I have had the rare opportunity in my life to be part of several revolutions and changes that have affected humanity. I have experienced the advent of computers, the Internet, and now also AI. There has been an incredible development in my field of numerical simulations. I enjoy watching the change and also being in constant contact with people who are really interesting, have a drive and bring new ideas. That's always fantastically enriching. When a person hides from the rest of the world, it does not benefit them or those around them. But if you have the opportunity to communicate – even with the best – it's something I can recommend to everyone.