The ability to extract the three-dimensional reality from two-dimensional images and to forward simulate potential consequences of our actions are fundamental aspects of human intelligence. “We are used to doing this effortlessly,” says Dr. Adam Kortylewski. “For machines, however, it is extremely difficult,” he continues. “In our everyday research, we see how far AI still is from human-level visual intelligence.” This tension drives his research. Specifically, he investigates how AI models create visual representations, and how robust these representations are—how they fail, how they can be manipulated, and how they can be made more reliable.
“I am particularly interested in the way generative AI models are changing our assumptions about visual representation learning,” he explains. “AI models no longer just learn to extract information from images—they can also generate them. This opens entirely new perspectives on visual intelligence.” His work sits at the intersection of machine learning, visual computing, and computer security. Machine learning provides the methodological foundation, while visual computing frames the central questions. “If we could understand and model visual intelligence at a human level, it would have enormous real-world implications,” he believes. From computer security, he takes a critical mindset to view AI as a real, fallible, and vulnerable system—not as a perfect black box.
Excellence and Ambition: Reasons for Joining CISPA
Kortylewski’s move from the Max Planck Institute for Informatics and the University of Freiburg to CISPA was a conscious decision. What tipped the balance was the combination of long-term research freedom, resources, and a critical mass of experts in machine learning and cybersecurity. “CISPA is scientifically excellent and ambitious at the same time. It invests in long-term questions and provides great resources,” he says. Understanding fundamental problems often takes time. What drives him as a researcher is the opportunity to ask fundamental questions. “This process of seeking answers is something I truly enjoy and what makes being a researcher the best job in the world,” he explains. “The feeling of discovering something new—no matter how small—and understanding the world a little better is incredibly motivating.”
Germany as a Location: Long-Term Perspectives and Intellectual Independence
Kortylewski has conducted research in several countries and is familiar with different scientific cultures. While he found the environment in the U.S. extremely dynamic—with people willing to take risks, launch projects quickly, and change direction if necessary—during his PhD studies in Switzerland, he found that the country represents for him the combination of excellent funding with high scientific standards, often in smaller teams with close industry ties. Germany is particularly attractive for him because it offers long-term perspectives and intellectual independence. “Researchers are not under constant pressure to chase the next grant. Their intellectual independence is well protected. This approach encourages researchers to invest in ideas that may take years to mature but can reshape entire fields.”
The Generative Intelligence Lab in Saarland
Currently, Kortylewski is moving his research group to CISPA: the Generative Intelligence Lab. Its goal is to investigate fundamental questions of artificial visual intelligence and better understand the interplay between image generation and image understanding. To realize his ambitious agenda, he is recruiting PhD students, whom he aims to support on their path toward becoming independent researchers through collaborative intellectual growth. Starting at CISPA is easier for him because he has already been living in Saarland for several years due to his prior affiliation with the Max Planck Institute for Informatics. This allows him to better integrate professional and personal life. “I find Saarland very friendly and calm, with open-minded people,” he says. “It’s an environment where you can focus well and feel at home quickly.”
A Clear Message: AI Should Support, Not Replace Humans
Kortylewski’s passion for understanding and improving artificial intelligence is reflected in his clear stance on the challenges and limitations of AI systems. “AI can accelerate scientific discovery, democratize access to knowledge, and automate monotonous or dangerous tasks,” he says. At the same time, he clearly identifies the challenges: biases in data, hard-to-interpret decisions, manipulability, and misuse. This is why developing systems that are reliable is so important to him. “AI should support human judgment, not replace it,” he emphasizes. “For that, we need frameworks where humans retain control over powerful but imperfect systems.”
