Dr. Tomasz Szandała from the Department of Technical Computer Science won a prestigious ESKAS Swiss Government Excellence Scholarships for young doctors . At the University of Lugano, he will conduct research on the interference of artificial intelligence in photos.

Swiss Government Excellence Scholarships are an elite Swiss government scholarships for young foreign scientists who have completed master's or doctoral studies and for foreign artists with a bachelor's degree. Only 200 such scholarships are awarded annually, including several dozen for doctors. Each of them is applied for by many young, ambitious people, both from Switzerland and from 180 other countries. The competition is huge. Many documents need to be prepared..

– Some people thought that I had little chance of getting a scholarship in Switzerland, especially a scholarship from the Swiss government – admits Dr. Tomasz Szandała. Despite their pessimism, he decided to compete for a prestigious scholarship and new development opportunities in an international group. And rightly so. He got into the program as a candidate from the first pool. – The scholarships are very prestigious. I heard that last year only four people from Poland received such a scholarship. I am even more happy as I will start a one-year postdoctoral internship in beautiful Lugano, on the border with Italy.

Our Faculty's employee moves to the south of Switzerland in the autumn. At the The University of Applied Sciences and Arts of Southern Switzerland together with Dr. Omran Ayoubem and Prof. Silvią Giordano Cremonese he will conduct research on recognizing photo manipulation using artificial intelligence.  

Our scientist wants to explore robust techniques behind the emergence of sophisticated methods for generating deepfakes, i.e. digital manipulations generated by artificial intelligence (AI) that use deep learning techniques to overlay or replace the likeness of one person with that of another person in images and videos.

– This is because content generated by artificial intelligence can be used for deceptive purposes, such as disinformation and its dissemination, attacks using social engineering, which can potentially undermine public trust, social cohesion and individual security – explains Dr. Tomasz Szandała. – Our team's research proposal aims to solve this problem by developing an AI-based deepfake detection system that is easy to explain and reliable. The system will not only identify potential deepfakes, but also provide explanations regarding the reasons behind its decision.

The methodological approach assumes the use of a diverse and inclusive data set and state-of-the-art technologies in the field of deep learning, such as convolutional neural networks or transformers known, for example, from ChataGPT.

– We also use depth analysis techniques, error analysis and explainable AI techniques – emphasizes Dr. Tomasz Szandała. An important goal is to create a comprehensive framework for evaluating our system that will include both quantitative (sensitivity, specificity) and qualitative (fidelity, interpretability, explainability, and fairness indicators) assessments.

The idea for such a research project appeared in the head of our Department of Technical Computer Science scientist while writing his Ph.D. thesis and researching how neural networks work. – I decided that I would be able to use this knowledge to recognize whether the neural network had manipulated an image – says Dr. Tomasz Szandała. Two events motivated him to pursue this goal.

– During the holidays last year, generative artificial intelligence began to create various photos. For example, the famous photo with the Pope in a white winter jacket or a leather biker jacket went viral around the world. There are also photos of politicians, celebrities and even ordinary people who may not have the means to defend their good name. I think that over time there will be more and more situations where photos that look authentic will be used as a tool to ridicule or discredit someone. As a tool, used e.g. by young people in schools, to ruin someone's life – notes our researcher and recalls an event from Spain. – During the holidays, twenty girls received photos generated using artificial intelligence, which were intended to discredit them in the eyes of their parents or school. It turned out that all these photos were manipulated. Someone admitted to doing it. But what if no one takes responsibility? You will simply have to prove that the photo is false in order to defend yourself. I hope my work will help with this. I believe I can cope with the current artificial intelligence. But I also know that in a few years the project will have to develop further to keep up with the development of AI.

r. Tomasz Szandała is gjoing to Lugano for a year, but he is not going to disappear from WUST. He wants to still teach his diploma students and run classes with students. Especially since he won't be teaching students in Switzerland at first.

– To be able to conduct classes there, you need to prove that you are actually a specialist in a given topic and have experience, but you also need to gain the recognition of your teammates and people with whom you cooperate – says the scientist. – So, at least for the first six months, I will not be able to teach in Lugano, but I hope that this will change in the summer semester. I would love to bring the local teaching methods to Poland.

Many people around Dr. Tomasz Szandała predict that he will stay in Switzerland for longer than a year. And he does not rule out longer cooperation with scientists from Lugano... – When applying for this project, I planned to create a team consisting of students from Wrocław and people from Switzerland. I know that several people there deal with similar topics – he says. – I want to create an international team based on the youngest possible people. Will I stay longer in Lugano? I don't know. For now, I'm glad I ended up there, because I've been learning Italian for several years, and Lugano is a beautiful region and the only canton where Dante's language is spoken ;-)