Alessandro Antonucci
Beyond the limits of AI
SUPSI Image Focus
Alessandro Antonucci, Associate Professor in Robust Probabilistic Artificial Intelligence at the Department of Innovative Technologies, retraces the path that led him from theoretical physics to AI. For over twenty years, he has been working on reliable intelligent systems, analysing their limits, safety and transparency, with the aim of making research outcomes concrete and applicable, including in the context of language models.
Where do you come from, what did you study, and what are you currently working on?
After completing a degree in theoretical physics in Milan, I intended to continue my studies with a PhD, not necessarily on the same topics, but with a strong mathematical focus. This led me, more than twenty years ago, to work in artificial intelligence at the Dalle Molle Institute for Artificial Intelligence Studies (IDSIA USI-SUPSI), an activity I continue to pursue today with the same enthusiasm. While at the time AI was an almost exotic discipline, largely unknown to most, it has now entered the everyday life of many people, even if it is still not widely known or fully understood.
What research topics do you focus on and what motivates you the most?
My work mainly focuses on “robust” AI, that is, how an intelligent system can react to external perturbations. This is extremely important to understand the “limits” of such systems, assess their safety, explain the outputs they produce, and evaluate the presence of possible biases.
In essence, I aim to improve the tools currently used in contemporary AI. I find this highly engaging both from a technical perspective, in terms of the mathematical and conceptual challenges involved, and in terms of its potential impact.
What has been the biggest challenge in your research path so far and what are the most significant results?
The challenge is always to combine fundamental research with applications. Each time I develop a new algorithm after a phase of theoretical work, the challenge is to make its impact practical and relevant.
I believe that some of my research on sets of distributions has had a significant impact in the field of robust AI. The most urgent objective is to apply these techniques to large language models (LLMs) in order to make them safer and more transparent.
Is there a particular personal or professional experience that has shaped your path?
In the early part of my career, I was entirely focused on research carried out individually. Over time, I have learned the importance of teamwork, the possibility of multiplying one’s potential and the impact of one’s work and, at the same time, I have begun to understand how important the role of teaching is, whether in the first year of a Bachelor’s programme or as a supervisor of a final-year PhD student, and how this is an integral part of the work of a good researcher and a source of satisfaction at least comparable to that derived from research.