AI systems already decide on bank credit, hiring and fraud detection. But how do we make sure those decisions are fair and can be explained to the people they affect? Sérgio Jesus tested, with real analysts rather than in a lab, how AI explanations shape human decisions, and showed that existing methods don't always protect against algorithmic discrimination. He also contributed to the largest open dataset for testing fairness in fraud detection systems, now used by researchers worldwide.

Sérgio Jesus is a researcher at Feedzai. His PhD thesis at the Faculty of Sciences of the University of Porto is precisely the work that has been awarded.

Fixing programming errors takes up a significant share of computer science teachers' time, even for simple, recurring mistakes among students who are just starting out. Pedro Orvalho created MENTOR, a system that automatically identifies the error in a student's code and delivers immediate, personalised feedback, without ever handing over the ready-made solution, so the student reaches the answer on their own. The result multiplies what both sides can do: the student learns faster and with more independence, getting an answer exactly when they need it, and the teacher gains back time to focus on the questions that genuinely require human attention, instead of repeating the same correction to dozens of students.

Pedro Orvalho completed his PhD at Instituto Superior Técnico and is now an MSCA Postdoctoral Fellow at IIIA-CSIC, in Barcelona.

How can two hospitals compare patient data without sharing confidential information? How can two companies detect and compare gender pay gaps without revealing individual salaries? Manuel Goulão developed multiparty computation protocols that let several entities compute joint results without any of them seeing the others' data, with applications ranging from market equilibrium to medical research. His work starts from the principle that privacy is a fundamental human right that technology should protect, not threaten. It was also built with the future in mind, designed to withstand attacks from quantum computers that will render today's encryption methods obsolete.

Manuel Goulão is now a researcher at INESC-ID, following a stint at the Okinawa Institute of Science and Technology, in Japan.

AI systems keep getting more capable, but they also demand ever more computing power, which translates into higher energy consumption and a growing environmental footprint. At the same time, they're becoming increasingly opaque, unable to explain why they reached a given decision. Gonçalo Correia developed a method that teaches these models to tell what's relevant from what isn't, instead of processing all information with equal weight. The result is lighter models that use less energy and are more transparent, able to show the people who use them, and the people affected by their decisions, the reasoning behind them. In simple terms, this means AI systems that are more sustainable and more trustworthy, at a time when we increasingly depend on them for decisions that affect us directly.

Gonçalo Correia is now Head of AI at Priberam and a guest lecturer at Instituto Superior Técnico, in Lisbon.