20 Ottobre 2026 | 14:00
DEIB, Aula 1A (Ed. 20)
Contatti: phd-inf@polimi.it
Sommario
On October 20th, 2026, at 2:00 pm Luca Morandini, PHD Student in Information Technology, will give a seminar on "The model worked, the system failed: Trustworthy AI for public institutions" in DEIB, 1A Room (Building 20).AI systems are increasingly used to support decisions about health, justice, and public safety. When they fail, the model is often not the root cause.
The failure comes from the data the model was trained on, how outputs are explained, and who is accountable for the result.
Drawing on recent literature, the seminar shows why technical solutions are necessary but not sufficient for trustworthy and interpretable AI.
Good performance and calibrated uncertainty estimates do not guarantee fairness or a meaningful explanation. Real-world cases are used to connect the technical side (data quality, uncertainty, interpretability and explainability) to its social consequences for the people affected. The seminar then discusses privacy and who is responsible under the EU framework (GDPR, AI Act). The talk argues that building trustworthy AI for public institutions means designing the whole system around the model, not only the model itself.