Igor Đurović, PhD: Four Jobs Annulled by Artificial Intelligence Result in Only One New One
Professor at the Faculty of Electrical Engineering, member of the Montenegrin Academy of Sciences and Arts, and one of the most cited Montenegrin scientists on Stanford’s list of the top 2% most influential researchers worldwide, Mr Igor Đurović, PhD, has been pushing the boundaries for decades in the fields of signal and image processing, artificial intelligence, chaos theory, and multimedia watermarking. His research has applications in areas that shape our future — from telecommunications and security to medical diagnostics and modern multimedia systems.
In an interview for the University of Montenegro website, he discusses current trends in science, the challenges posed by new technologies, and the importance of research that connects Montenegro to the global scientific community.
UMNE PR: If you had to single out one research question that currently most occupies the global scientific community in the field of signal and image processing, what would it be and why?
Đurović, PhD: “Under the hood of artificial intelligence tools that have emerged in recent years are advanced optimization algorithms. These algorithms originated in digital signal processing or are shared between these two fields. Therefore, the answer to your question is straightforward: the synergy between digital signal processing and artificial intelligence is the trend. There are two levels of this ‘collaboration’: one is fundamental — further improvement of learning systems and their optimization; the other is applicative — the use of artificial intelligence systems, primarily related to various multimedia signals (image, sound, video). If I had to highlight one specific area where I expect dramatic progress, it would be data fusion from different sources. The algorithms already exist, and we are just waiting for other pieces of the puzzle to align so that the number and accessibility of such tools can increase significantly.”
UMNE PR: Artificial intelligence is already shaping the way we live and communicate. Where do you see its greatest potential, and where do you see the greatest risks?
Đurović, PhD: “Artificial intelligence will transform almost all human activities. Only those tasks that are too expensive to automate or represent the highest achievements of human labour will be less affected. Whether this is ultimately good or bad, time will tell. The most immediate risk concerns the labour market. At the moment, for every four jobs that artificial intelligence eliminates, only one new job is created. I am afraid this trend will continue.”
UMNE PR: Your research connects chaos theory, statistics, AI, and image processing. Could you explain how all of this works using a simple example?
Đurović, PhD: My most significant results over the past decade are related to nonparametric and parametric estimation of non-stationary signals. These types of signals appear in speech processing, biomedical phenomena, radar, and sonar applications. Without false modesty, some of the results we have achieved rank among the best in this field. The other topics you mentioned are either tools that I use and apply in my research or application areas of the techniques I have developed for nonparametric and parametric estimation and time-frequency signal analysis.
As I have already said, artificial intelligence is the leading trend in modern science. My research involving artificial intelligence can be divided into three main areas: parameter estimation, multimedia signal authentication, and time series analysis. Although we have published several papers related to parameter estimation of signals using artificial intelligence tools, I must admit that the results are still inferior to classical techniques. The same holds true for other researchers as well. The reasons for this are not yet fully understood, but significant efforts are being made to improve the performance of artificial intelligence tools in this area.
Regarding the application of AI in multimedia signal authentication, after several successful studies involving artificial intelligence in digital watermarking, two of my PhD students founded a tech start-up called Deepmark. I am currently acting as a consultant for this start-up, which has the potential to become a success story not only in Montenegro but also within the regional innovation ecosystem.
Finally, we have ventured into applying artificial intelligence to the analysis and prediction of time series, particularly for forecasting electricity consumption within the power system. This research was conducted at the Montenegrin Academy of Sciences and Arts with the support of Elektroprivreda AD Nikšić. The results are outstanding and have been confirmed at the highest internationally verified level. To be honest, I am surprised by how successful artificial intelligence has been in this application compared to classical statistical tools.
Progress of Deepfake Videos and Fake images is Dramatic
UMNE PR: Deepfake videos and fake images are becoming increasingly realistic, making it harder to distinguish between what is real and what is artificially created. Can science keep up with these challenges, and how close are we to developing technologies capable of detecting every manipulation?
Đurović, PhD: The progress of deepfake techniques is dramatic. Currently, publicly available tools mainly fail only when it comes to shadows because they struggle to adapt to situations where different parts of objects in an image are illuminated by various light sources, resulting in multiple shadows. It is highly likely that this problem will be solved in the next few years or even months. At that point, visual inspection, even by an expert, would no longer be able to detect traces of image manipulation.
With videos and images, there is a fortunate circumstance: our semiconductor cameras introduce invisible errors, which means that deepfakes can be detected using advanced techniques capable of determining whether something was captured by a real sensor or generated by a computer based on invisible content. There is hope that deepfake systems will still need considerable time to learn how to fake this invisible image content.
The situation is worse when it comes to audio. Sound is, by nature, mechanical, and there is no “inaudible” content resulting from the acquisition process, which makes it much easier to fake. This is one of the reasons behind the creation of the start-up Deepmark, which is currently primarily focused on audio and speech authentication. In short, technology has so far proven to be much better at destruction than protection, and when it comes to defence, we are forced to rely once again on the same tools: deep machine learning and artificial intelligence.
Montenegro on a Global Scientific Stage: Key Support for Young Researchers
UMNE PR: As a scientist whose work is featured on Stanford’s list of the top 2% most cited researchers in the world, how challenging is it to conduct high-level research in Montenegro while remaining present on the global scientific stage?
Đurović, PhD: Access to scientific information, opportunities for networking, and academic mobility have never been better. Let us be honest, being a scientist in Montenegro is much easier today than it has ever been before. We should take advantage of the current situation because I fear that the ongoing “brain drain” trend, driven by large tech companies, has the potential to fragment science and gradually make it less open.
Is it possible to improve things? Certainly. What seems easiest to implement, compared to other measures, is increasing the number of research positions, reducing teaching loads for young scientists, and applying stronger pressure on scientific and higher education institutions to secure projects that allow for the engagement of researchers.
UMNE PR: If you were a student today, which fields and skills would you choose to study to prepare yourself for the science of the future?
Đurović, PhD: During my pre-university education, I was drawn to mathematics and its real-world applications, which strongly influenced my choice of studies. That choice would not have changed—until about three or four years ago. Today, we are aware that a profound revolution has taken place, driven by artificial intelligence and its practical applications, leading to tectonic shifts that we are only beginning to experience.
I would say that students graduating from undergraduate studies will need far more specialized knowledge, covering multiple disciplines, and that certain postgraduate programs, especially those with a clear focus, will need to be moved to the undergraduate level. This is a trend that has only just begun at leading universities worldwide, and we will have to adapt to it.
Therefore, I cannot give young people a definitive recommendation on which field they should choose to build their education and career for the coming decades.
UMNE PR: What are you and your team currently working on that could change the way we view signal and image processing in the coming years?
Đurović, PhD: I am trying to maintain the pace of scientific work while aligning my research within the framework of new scientific developments. I have already mentioned three directions of our research: parametric and nonparametric estimation, digital watermarking and multimedia data authentication, and time series estimation — all aimed at advancing the development and application of artificial intelligence.
I am also interested in exploring the concept of framing artificial intelligence systems as “black boxes” that may approach the theoretical limits predicted by information theory.