Following a successful first session, the second INESC-ID Distinguished Lecture was held last week, on 2 July, at Instituto Superior Técnico featuring Professor Ricardo Baeza-Yates from KTH Royal Institute of Technology, Universitat Pompeu Fabra and University of Chile.

Entitled “Evaluating ML: When will we stop fooling ourselves?”, the talk reviewed how Machine Learning (ML) evaluation often relies on average success measures such as accuracy and highlighted their limitations. According to the speaker, these included models may perform well on easy instances but poorly on difficult ones, errors rarely have equal impact, and optimising overall success can fail to reduce critical mistakes. After outlining these drawbacks, the session programme then examined the underlying issues and presented solutions to address them.

Rodrigo Rodrigues, INESC-ID Scientific Council Coordinator and researcher in Distributed, Parallel and Secure Systems stated that “the session addresses one of the most current topics in our research, namely Machine Learning Algorithms, which are at the heart of the AI revolutions, and their limitations. Within this area, the importance of the lecture relates to its focus on the issue of assessing the reliability and safety of using AI in systems and tasks critical to our society”.

© 2026 INESC-ID. When sharing or adapting this article, credit INESC-ID with a link to the original source.

Image: © 2021 INESC-ID