In the second ELOQUENCE WebCafé session, titled “Trustworthy AI: Why Explainability Matters,” the discussion focused on one of the most important questions in today’s AI landscape: how can we understand, control, and trust the systems that increasingly shape the way we communicate, work, and make decisions?
The session welcomed Sergio Burdisso, researcher at the Idiap Research Institute in Switzerland and a member of the ELOQUENCE project team. His work focuses on explainability, context-aware interaction, and the development of AI systems that are not only powerful, but also transparent, reliable, and human-centred.
Sergio began by sharing his personal journey into artificial intelligence and language technologies. His interest in computers started early, driven by curiosity and the desire to understand how things work. This eventually led him to study computer science, pursue a PhD, and continue working in language technologies — a field he described as especially fascinating because it brings together computer science, psychology, mathematics, and language.
A central theme of the conversation was the difference between AI in research and AI in industry. Sergio noted that while companies have often focused on whether a system works and delivers value, the growing presence of AI in everyday life has made trustworthiness a much more urgent concern. Today, AI systems are no longer used only by researchers or technical experts. They are increasingly part of tools and services used by the wider public, which means their societal impact must be taken seriously.
Within the ELOQUENCE project, Sergio’s work focuses on explainability and context-aware interaction, especially in high-risk scenarios. In such contexts, AI systems may influence decisions that affect people’s lives. This makes it essential to understand not only what output a system produces, but also why it produced it. Without explanations, AI remains a black box: users provide input, receive output, but cannot assess whether the system’s reasoning is valid, biased, or reliable.
The discussion also explored what it means to make AI more human-centred. For Sergio, this involves two important directions: allowing human feedback to improve AI systems, and enabling experts to inject knowledge into the models. Instead of letting AI generate responses freely without control, the goal is to guide systems through structured steps designed by human experts. This makes the system easier to understand, easier to evaluate, and more aligned with real-world needs.
Sergio also spoke about his involvement in the Jelinek Summer Workshop 2025, where researchers from around the world will work intensively on controlling, evaluating, and understanding large language models. The outcomes are expected to contribute directly to the goals of ELOQUENCE, particularly in relation to explainability, interpretability, and trustworthy AI.
The session concluded with an optimistic view of the future. While modern AI systems are complex and often difficult to interpret, growing attention to regulation, transparency, and open research is helping the field move in the right direction. Explainability is no longer a secondary concern, it is becoming essential for building AI systems that people can understand, trust, and use responsibly.
Listen to the full episode here.
