Prof. Dr. Dubravko Ćulibrk, director of the Institute for Artificial Intelligence of Serbia - In 2030, the ICT sector may export 12 billion USD, radiologists will use AI, but we will still drive cars by ourselves

And what does Prof. Dubravko Ćulibrk, director of the Institute for Artificial Intelligence of Serbia, predict?

The information and communication technology sector in Serbia is growing at an annual rate of over 20%, and we can expect that the ICT sector will export USD 12 billion in 2030, reveals Ćulibrk in an interview for eKapija.

As he says, according to the index of readiness of the state for artificial intelligence by the company Oxford Insights, Serbia ranks 57th among 193 ranked countries. We note the fastest development of AI in the financial industry - banks already use AI for creditworthiness assessment and fraud prevention - followed by ICT, healthcare and e-commerce.

Ćulibrk notes that more and more domestic companies recognize the potential of AI technologies, but that SMEs do not believe in AI systems even though they can profit the most from digital transformation.

The solution to many challenges that AI brings is in regulation, and he hopes legislation on artificial intelligence will be passed by the end of next year.

And no, Ćulibrk does not run away from the forecast from the introduction. What does he think about it, but also what artificial intelligence tools he uses in his daily work, and whether he is afraid that AI will lead to "everyone living in his own world", find out in the following lines.

eKapija: Artificial intelligence (AI) will contribute USD 19.9 trillion to the global economy by 2030 and boost 3.5% of the global GDP in 2030, the International Data Corporation (IDC) recently announced. Do we have similar estimates for Serbia?

- As far as I am aware, specific assessments for Serbia regarding AI have not been published, but it seems to me that AI can be viewed in the context of the ICT sector, which in Serbia is growing at an annual rate of more than 20% and in 2022 accounted for more than 5% of the GDP of Serbia. Exports from this sector are expected to reach 4 billion EUR by the end of this year. The ICT sector is certainly the sector that integrates AI solutions into its operations the fastest and we can assume that its further growth and development will be significantly supported by artificial intelligence. In this sense, it is to be expected that Serbia will certainly achieve this global estimate that 3.5% will be stimulated by VI in 2030, and a bit of simple math tells us that we can expect the ICT sector to generate some 12 billion USD in 2030.

In the sectors that, as the "locomotives" of growth in our country, have been ahead of ICT lately (construction, wholesale and retail, then mining), I think we can expect a considerable rapid and positive impact of AI on trade, while the introduction of AI-based solutions in the remaining two sectors will need to go hand in hand with digital transformation and it will take more time. In any case, by 2030, the AI should enable a part of the income and growth of these sectors as well, so an estimate based on the ICT sector alone should be conservative in that sense.

eKapija: How do you evaluate the development of AI in Serbia compared to the world and the region? How much do companies, state institutions, individuals apply it?

- The development of artificial intelligence in Serbia is in an upward phase, but we are still lagging behind the leading world centers, such as the USA, Canada and China. According to Oxford Insights` AI Country Readiness Index, we are ranked 57th globally out of 193 countries ranked.

In the regional framework, according to the same ranking, Serbia lags only behind Slovenia, Hungary and Bulgaria, which have invested considerably more in this development so far. In relation to the entire Eastern Europe region, our country is leading the way in terms of vision, ethics and regulation and the so-called digital capacity. Based on data from 2023, we lag behind significantly in terms of infrastructure and data availability. To a lesser extent, also with respect to the maturity of our system in this domain. I believe that with the formation of the Artificial Intelligence Council, with the planned investments in infrastructure and the passing of the law on AI, we will improve our position in the coming years.

More and more domestic companies recognize the potential of AI technologies and implement them in the areas of analytics, business process automation and customer support, thereby increasing efficiency and competitiveness on the market. At the same time, state institutions are starting to introduce AI into their systems, although a more intensive introduction of these solutions is needed in order to achieve a greater impact on public services.

"Application of AI in Mammography" project (Photo: Institut za veštačku inteligenciju)

From the point of view of our institute, as an example we can cite the cooperation with the Institute of Radiology and Oncology of Serbia on the "Application of AI in Mammography" project. Also, the Institute cooperates successfully with the companies "Telekom Srbija" and "Elektroprivreda Srbije", and from the industrial partners, I would single out the cooperation with several pharmaceutical companies in the framework of which work was done on the detection of rare diseases, therapeutic peptide and small molecule design, etc.

When it comes to individuals, it`s hard for me to estimate, and I couldn`t find any statistics, how many users of services that are primarily based on AI, like ChatGPT, but according to DIGITAL 2024 report, in Serbia there are 5 million users of various social networks (more than 70% of the population). Given the level of AI integration in these platforms, we can safely say that all of these people are using AI. If they are aware about what they are using and what the limitations of these technologies are, that is again another question.

eKapija: In which area is the application of artificial intelligence in Serbia developing first - in the financial industry, healthcare, trade... In which industries is the process difficult and why?

- The fastest development of AI is recorded in the financial industry, ICT, healthcare and e-commerce. Banks use AI to assess creditworthiness and prevent fraud. In healthcare, machine vision algorithms, as a subfield of AI, are now already standardly used for medical image analysis and diagnostic support. AI algorithms are present in e-commerce for personalizing the offer and optimizing the stock.

When it comes to industrial production, the digital transformation is progressing more slowly due to the lower level of digitization and the lack of specialized staff. The implementation of new technologies requires not only the modernization of equipment, but also the development of employees` skills, which is currently a challenge. Without adequate support in education and training of experts, the process remains fragmented and slowed down.

eKapija: What are the key obstacles for AI development in Serbia?

- Considering the AI ​​Readiness Index, the biggest obstacles are the infrastructure and availability of data, which is closely related to digital transformation in all areas of our society.

From the standpoint of our institute and the accomplishment of our mission, the biggest obstacles to the development of AI in Serbia include the lack of highly qualified experts and limited financial resources for research and development. Due to the global demand for experts in the field of artificial intelligence, Serbia is facing challenges of retaining talents, while many young professionals decide on foreign careers. Limited budgets further slow the development of research projects, especially in small and medium-sized enterprises as they do not have the capacity to invest independently in AI technologies.

Apart from financial challenges, the regulatory obstacles make implementation of new technologies difficult. The legislative framework is not yet fully adapted to the rapid changes in the field of AI, which creates uncertainty for companies and research institutions developing innovative solutions.

Another important barrier is insufficient awareness and trust in AI systems, particularly among small and medium-sized enterprises, even though they can profit the most from digital transformation. Without adequately understanding the potential of these technologies, many business entities remain in the status quo and thus miss the opportunity to increase efficiency and productivity.

Team of the Institute of Artificial Intelligence of Serbia (Photo: Nebojša Rudić)

eKapija: As a professor at the Faculty of Technical Sciences, how do you evaluate the knowledge and interest of students in this field?

- From my personal point of view, I would say that students` interest in AI is on the rise, as, for example, at the Faculty of Technical Sciences, as well as at the University of Novi Sad, we are witnessing a growing interest in master`s and PhD programs dealing with AI and data science. Even on study courses where these technologies are studied within elective subjects, students are increasingly choosing courses that deal with this subject.

I don`t mean to be harsh, but I would say the knowledge is pretty limited and fragmented. Last week, I asked my students in the Master`s degree in Information Systems Engineering if they know who Geoffrey Hinton is and more than half didn`t know about the man who is called the "godfather of AI" and who received both the Turing and Nobel prizes for his services in this field. We didn`t pass well with the question about what the Turing Award is and who Alan Turing was, either.

On the other hand, practically everyone had the opportunity during their basic studies, graduated in various fields of the Faculty of Technical Sciences, not only to learn about artificial intelligence systems, but also to train deep neural networks, which are the basis of practically all modern AI systems. This is quite enough to prepare them for a successful career in this field by the end of their master`s studies, but perhaps we should also think about an entire programs of basic studies in the AI field.

As ever in engineering, special attention should be paid to practical trainings and opportunities for participation in real projects in order for students to acquire specific knowledge and skills that the market requires. But that has always been the approach at the Faculty of Technical Sciences in Novi Sad.

eKapija: If you ask ChatGPT, by 2030, it will be common for AI to analyze our MRI scan and diagnose disease, that we are questioned by AI interviewers at job interviews, to eat lab-grown meat, while fully autonomous vehicles on the road will be common between 2030 and 2040. The same deadlines according to ChatGPT apply to Serbia, except for lab- meat, which may be available in 2035, "but its entry into the mass market will probably require several more years of development and adaptation". Do you agree with the AI`s assessment, or do you still see something that the AI didn`t take into account?

- Basically, that answer is the standard level of accuracy for a generative model like GPT.

AI is often already used as an assistive technology in radiology and algorithms will become more and more efficient. Making the diagnosis itself will, most likely, have to remain in the domain of man. To paraphrase a colleague: "AI will not replace radiologists. Radiologists who use AI will replace radiologists who do not." Fully autonomous diagnostics is, it seems to me, in the far future and reserved only for the scenario when it will be absolutely necessary. Similarly, the AI ​​interviewer will need to be under some human supervision. When it comes to fully autonomous vehicles, I`m afraid GPT doesn`t consider the value of a typical fully autonomous vehicle, and, in order to make this prediction come true, most likely we need to abandon the paradigm of owning and driving our own car. I don`t think that this is going to happen in 10-15 years. What I think will actually happen by 2030 may not sound so groundbreaking, but I expect that by 2030, the way we access information will change drastically, as well as education, which will have far-reaching consequences. I really don`t know about lab meat.

I have to say that as a huge fan of science fiction, I advise readers to take everything I say with a grain of salt. These days I am reading a collection of stories published by the great Arthur Clarke in 1957, in which he predicts satellite television, but instead of satellites like we`ve always used, everything is based on a cluster of manned space stations. History rarely cares about what people and AI imagine the future to be.

eKapija: What artificial intelligence tools do you use in your work and in your daily life?

- It`s a bit like a shoemaker and his shoes.

I focus more on developing certain technologies, rather than their use, and I am also aware of the limitations of different technologies, so for example I will use Perplexity (an alternative to ChatGPT) to generate a content proposal for a popular lecture that I should give, but then in most cases I will change it that much that the question arises whether it is more efficient to have made it myself from the start. What really helps me in this kind of task are models like DALL-E which I like to use for generating illustrations.

From "classic" applications of artificial intelligence, music recommendation systems are definitely my favorite which I have been listening to for almost 20 years on the internet platform which is now a classic itself.

As far as the development itself is concerned, as someone who primarily deals with computer vision, I started with C++ and OpenCV, switched to Python just over a decade ago, using Caffe first, then TensorFlow, and lately I`ve been using Pytorch mostly.

Unfortunately, I don`t have much time to do coding anymore, but younger colleagues use chatbots extensively for inspiration and debugging the code. As for me, I hope that this year, in collaboration with the students, we will develop the first chat bot that should make it easier for students to go through the material on a new course. It is not difficult at all in general, but we would like to develop it ourselves relying on open-source technologies and locally hosted, so that part will be a bit more demanding.

eKapija: In which area do you not want to see the application and development of AI and why?

- I would not define it at the level of the field, but of certain applications, as it is defined in the AI ​​law of the EU, when talking about prohibited AI systems.

These are e.g. systems that manipulate human decisions and exploit their weaknesses, systems that evaluate and classify people based on their behavior and personality traits, as well as systems that assess the risk that a person will commit a criminal offense in the future. There are also systems of mass surveillance in public places. except when it comes to really specific situations. I don`t think I need to explain why.

AI technologies have a huge potential to improve various sectors of society on the condition that we use them responsibly and ethically. Serbia already adopted the Ethical Guidelines for the Development, Application and Use of Reliable and Responsible Artificial Intelligence in March 2023. but not binding. I hope for the adoption of the law on AI by the end of next year.

eKapija: The benefits of AI are clear, but it is becoming increasingly clear that AI systems, like ChatGPT, are not as unbiased and objective as their creators want to make them out to be. Algorithms on social networks are already dictating content in accordance with the values ​​and political determinations of the owners, while, on the other hand, we tend to adapt content on the Internet to the values, beliefs and needs of users. As a user, what scares you the most - the presence of a censor who determines what information we will receive, which we are already familiar with to some extent, or the new principle "everyone lives in his own world"? What are its effects on society, will it destroy empathy, communication, the desire to meet and understand the other and different?

- I would say that they are just two sides of the same coin, i.e. "systems that manipulate human decisions" as I called them in the previous answer, and they scare me too. What I didn`t say there is that, unfortunately, that application has been known for a long time, has been used and is still being used. What Cambridge Analytica did was, in my opinion, the worst thing which could be done with advanced algorithms of the era.

I am neither a sociologist, nor a psychologist, but it seems to me that when everyone only consumes information that confirms their views, it probably reduces empathy and willingness to understand other perspectives. The solution to these challenges can be found, above all, in regulation, and then also in the transparency of the algorithms and the raising of media literacy among users in order to be aware of the way in which information is marketed. Also, the promotion of critical thinking and active encouragement of dialogue among different groups can contribute to better social cohesion, because AI systems must be designed and used to increase understanding between humans, rather than further reducing it.

Marija Dedić