News and events archive

Vít Nováček from the Faculty of Informatics at Masaryk University and his team investigate how to identify meaningful connections within the vast amounts of healthcare information available – connections that can support diagnosis, treatment, and ongoing patient care. How can AI help us cope with information overload? Why is it only a good assistant when we understand both its capabilities and its limitations? And does studying computer science still make sense in the age of AI?

Assoc. Prof. Vít Nováček, PhD, works at the Department of Machine Learning and Data Processing at the Faculty of Informatics, Masaryk University. His research team, the Discovery Informatics Group, investigates how fields such as artificial intelligence and machine learning can help address information overload, for example in the life sciences and news data. In this interview, he shares insights from applied medical informatics and discusses the broader use of AI tools for working with information more effectively.

You lead the Discovery Informatics Group at the Faculty of Informatics. What exactly do you do, and what is “discovery informatics”?

Discovery informatics is a subfield of computer science that generally focuses on techniques for managing and analysing data generated by research, often in areas such as biology, pharmacology, and medicine. In our group, we specialise in “making sense” of healthcare data. We work extensively with textual medical records, but also with structured data from medical research and clinical practice, and recently we have also begun cautiously experimenting with genomic data. We try to dig something useful out of all of this – something that could help healthcare professionals or patients.

One example is developing models to predict the risk of a second cancer after someone has recovered from their first. We also enable people with cancer to access personalised, intuitively presented information about their disease and treatment. This helps them gain greater control over their “patient journey” and, more broadly, their lives.

Could you give us an example of some specific projects?

Sure. We are working on personalised information for people with cancer as part of the EMPOWER project, funded by the Czech Ministry of Health. Among a number of other projects we are involved in together with the Masaryk Memorial Cancer Institute in Brno, IDEA4RC is also worth mentioning. It is a major international project funded by the European Commission that analyses data on rare cancers using machine-learning models operating across 11 hospitals throughout Europe.

On your research group’s website, you identify information overload as one of the major challenges of our time. Where do you encounter it most clearly in your work?

Pretty much everywhere. I think we all have some sense that there is far more information capable of significantly affecting our everyday decisions than we can possibly process in the time available to make those decisions. In ordinary life, that may not be such a big problem. If we overlook something when choosing a new car, we might grumble about it for a while; the car will still run, and eventually we will get used to it. But what if an oncologist overlooks a crucial piece of information when designing an experimental treatment for a patient with an aggressive tumour? The consequences are likely to be rather more serious than realising, a week after buying a shiny new Kodiaq TDI, that you actually hate the smell of diesel.

Your research brings together artificial intelligence, machine learning, knowledge graphs, large language models, and unstructured text processing. What connects these areas, and which is the most important to you today?

At the technical level, there is sometimes a surprising number of unexpected similarities and connections related to the historical development of these disciplines. But what I find more important is that they allow us to connect different pieces of the information puzzle relevant to the problems we work on.

Probably the most important point is that whatever AI technique we use, it is more or less useless until we can reliably estimate how it will perform in practice. At least in medical informatics, that remains an open challenge shared by all the areas you mentioned.

When AI Enters Medicine

You have been working in medical informatics for a long time. So how well is AI actually performing in practice?

In some areas, extremely well. In radiology, for example, various AI techniques for image processing have long been incorporated directly into diagnostic and treatment equipment, or into hospital information system modules used to process radiological data.

Another good example at the intersection of biology, pharmacology, and medicine is AlphaFold, DeepMind’s groundbreaking tool for predicting the three-dimensional structure of proteins. This application of AI has already contributed to the development of new drugs and biologically active molecules that genuinely help people. And, with a little exaggeration, AlphaFold also effectively dropped a Nobel Prize in Chemistry into the laps of a bunch of computer scientists, which I personally think is amazing.

That certainly does not mean, however, that AI is some miraculous solution to every problem in medicine. So far, success can typically be achieved only in clearly defined cases where clinical challenges can be reasonably translated into computer science problems to which an AI technique can then be applied. And even then, it is far from straightforward. It takes enormous effort and patience from people across different disciplines, and sometimes it takes them years just to learn how to communicate with one another effectively.

What makes the clinical environment challenging for computer scientists, and what makes it attractive from a research perspective?

One of the main challenges is precisely that medically interesting and important problems are difficult to formulate as computer science problems. It requires a great deal of work and highly motivated people from many different disciplines – not only computer science and medicine, but often psychology, biology, pharmacology, economics, law, and ethics as well. And those people have to be willing to understand one another, which is not something you can simply take for granted.

Robust clinical validation of AI goes hand in hand with this, and that is still largely uncharted territory. But those very challenges are also what make the whole field so fascinating.

Another nice thing is that the bar is often relatively low. Many problems in medicine still have no solution at all, so even a fairly experimental AI prototype can be genuinely useful to healthcare professionals or patients. And it is great to be part of that.

AI Needs to Be Smart – and Safe

The Onkorádce application appears in the AIcope and EMPOWER projects you are involved in, serving as a guide for patients during and after treatment. Could you explain what Onkorádce can – or should – do?

It should provide people with clinically validated information that is directly relevant to them and that they not only want, but also need to know. Perhaps even more importantly, the system must not give them information that they definitely should not receive at a particular moment – for example, something that could put their health at risk if they misinterpreted information about their condition and consequently refused an effective treatment.

Onkorádce also needs to be able to determine precisely when healthcare professionals should become involved in the process of informing patients, and how to make that involvement as efficient as possible for both sides. These are all things that tools such as ChatGPT cannot do, whereas we are able to address them thanks to our unique team, data, and interdisciplinary expertise.

What requirements must an AI system meet if it is to communicate sensitive information in oncology?

Well, that is precisely what nobody really knows yet. For medicines and medical devices – whether we are talking about a simple thermometer or a sophisticated linear accelerator used in radiotherapy – there are clear rules based not only on science, but also on legal, ethical, and all sorts of other analyses.

But for medical decision support using contemporary AI technologies – primarily language models – no comparable regulatory framework currently exists. In addition to applied research, however, we focus heavily on robust clinical validation and on designing safety mechanisms that reduce the risk of adverse consequences when our tools are used in medical practice. So I believe that once the regulatory landscape begins to move forward, we will be among those who have been approaching the issue properly from the outset.

Alongside computer science, you also studied political science and media studies. How has that background influenced the way you think about technology and artificial intelligence?

Not significantly in my day-to-day work. But perhaps journalism helps me talk about our research with people who do not necessarily understand all the technical details. And the fragments I remember from political philosophy may help me think about AI as a tool that is not merely technological, but also profoundly political and connected with power, with all the possible consequences that entails. And that is probably quite useful.

Does that background also make you look differently at data, texts, and the credibility of information?

Probably not differently – or rather, not differently from the way we should ideally all approach data and information: by constantly questioning not only what the outside world throws at us, but also what it does to us internally, so that we can continually move as close to the truth as possible.

Valuable Experience Abroad

You completed your PhD at the Digital Enterprise Research Institute (DERI) at the National University of Ireland Galway and subsequently held other research positions in Ireland. You also have experience from the United States. What did this extensive international experience mean for you professionally?

It gave me the opportunity to see how both science and the transfer of research into practice are done in different countries that, demonstrably, are not doing a bad job of either. It also gave me the chance to meet many excellent scientists and generally fascinating people from all corners of the world.

Not that this in itself necessarily advances you directly as a scientist. But realising that my personal perspective may represent only a very small slice of all the possible ways of looking at the world is, I think, priceless.

How would you compare the American, Irish, and Czech research environments – for example, in terms of openness to interdisciplinary projects, links between academia and industry, or support for early-career researchers?

That kind of comparison is fairly difficult. Each country starts from different circumstances, has a different history, and has different financial and organisational possibilities. And things change considerably over time.

The United States, for example, has recently become a rather different country – not only in science – from what it was about 15 years ago. People used to happily flock to American research institutions; now they are more likely to be leaving them, unless they are simply forced out.

Generally speaking, I would say that during the 14 years I spent abroad, the Czech Republic made considerable progress in terms of its openness to the international community and its support for excellent research. But there is certainly still room for improvement, for example when it comes to connecting academia with practice.

Ireland, for instance, has a dedicated agency, Enterprise Ireland, for this socially important agenda. It has a well-designed system of financial and personnel support for people who want to transfer their research into practice, for example by establishing a start-up. I still feel that the Czech Republic somewhat lacks this kind of systematic, stable, and transparently structured approach.

From Medicine to News Analytics

You also collaborate with Newsmatics, a member of the Faculty of Informatics’ Association of Industrial Partners. The company uses AI to analyse large volumes of online news and transform them into structured data on selected topics, trends, and the credibility of information. How does working with textual data in healthcare differ from news analytics?

The work is substantially different – the challenges and the way of thinking about them are completely different. I see it as a bit of a break from our group’s main research agenda, and it occasionally lets me reminisce about my rather entertaining late-noughties humanities studies. 😊

On the other hand, text-processing techniques transfer very well across different domains. So methods we use in healthcare can often be applied relatively straightforwardly when working with Newsmatics as well.

AI in Science

You also serve as an AI consultant to the Czech Science Foundation. How do you think artificial intelligence is changing the research environment – from preparing grant proposals and evaluating research to questions surrounding the responsible use of AI in science?

It is probably still too early to draw any firm conclusions. On behalf of the Czech Science Foundation, I am involved in a Science Europe working group that focuses on aligning approaches to the use of AI in research funding across Europe.

What we are seeing so far is that quite a lot is happening – some funding agencies have been experimenting for some time with more or less restrictive approaches to using AI both in writing and evaluating research proposals – but it is not yet possible to identify clear trends or define general policies.

That will have to become clearer soon, though, because obviously nobody is going to “switch off” AI in this area either. As it happens, I am sitting on a train to Budapest right now, on my way to a meeting of that Science Europe working group, so perhaps I will be a little wiser in a few hours.

You mention on LinkedIn that you are also a diving instructor. Does diving help you switch off from the world of data, or do ideas actually come to you underwater?

I have not done much diving since returning from Ireland. Once you fall in love with the rugged but beautiful Atlantic coast – where you can, for example, make friends with a wild dolphin – Czech quarries and lakes are not quite as exciting, although I still very much enjoy simply swimming in them.

But I will always love water in general. Not only because it is an element where I really cannot take a computer with me even if I wanted to, but also because it is an environment where, by its very nature, you can simply float freely. And then it does not matter all that much whether you are swimming above the tiles of a pool or out in the “big blue” with some amazing marine creature by your side.

Computer Science in the Age of AI

What question about AI would you like to be asked?

Do androids dream of electric sheep? I think that is a pretty great way to start a conversation about “AI with a broader perspective”.

And what would your answer be? Should we be afraid of AI, or does it still lack the necessary empathy? 😊

I would not be afraid of AI. It is simply a fairly powerful tool, but nothing more – at least for now – much like many other things humanity has come up with over the past few hundred thousand years: fire, penicillin, nuclear energy, space travel, gene editing, you name it.

What I am more afraid of is how humanity as a whole will respond to everything AI may come to mean for us. It would actually be enough if, as a species, we followed the maxim Wil Wheaton popularised at an online gaming conference almost 20 years ago: don’t be a dick.

Maybe we will get there one day, although things do not exactly look that way at the moment. I know, I have always been a bit of a dreamer. But so far, that has worked out reasonably well for me on a personal level, so hopefully I am not completely off the mark.

And what would you say to secondary-school students wondering whether studying computer science still makes sense in the age of AI?

I can offer a metaphor: think of AI as a chainsaw. You can certainly cut a lot of wood with it. But how do I know whether the wood has been cut properly? And if I am not exactly a professional lumberjack, I could quite easily use that big, unruly saw to cut off something other than wood – something I might later miss.

And even if I manage not to cut off anything important, it might not occur to me that a proper axe is actually better for splitting the logs I have just cut. Or perhaps a log splitter, if I can afford one. The saw might also break – what do I do then? And what if, instead of wood, I suddenly need to work with stone?

Sure, I might have a very smart saw that can talk and is happy to advise me on everything. But how do I know when I can trust it? And why should a saw be advising me on breaking rocks in the first place? …

In any case, I believe studying computer science makes more sense in the age of AI than ever before – especially at universities that try to place the subject in a broader context, which is something we at the Faculty of Informatics at Masaryk University are, I hope, doing quite successfully.

Thank you for the interview, and I wish you every success in your future work.

Author: Marta Vrlová
Photo: David Pařík