Researcher uncovers weaknesses in AI systems (#16)

Show notes

Sandra Casalini speaks with Maura Pintor, Assistant Professor at the PRA Laboratory of the University of Cagliari, about AI security. Pintor emphasizes the importance of proactive security testing of AI, where attacks are modeled in advance. She criticizes the fact that misjudgments and confirmation biases hinder progress, as researchers often overestimate how robust their models are. She also sees problems arising from the rapid development of new models and generative AI, which could compromise data quality. Nevertheless, she remains optimistic because progress is being made in automated testing.