Artificial Intelligence has proven its enormous capabilities over the past few years. In this talk, I demonstrate that these systems can be tricked into making catastrophic errors relatively easily. I present part of my research work that shows how today’s defense mechanisms have negative side effects on the performance of AI that make them unusable in practice.

Mathias Lechner is a Postdoctoral Research Associate at the Computer Science and Artificial Intelligence Laboratory (CSAIL) at the Massachusetts Institute of Technology (MIT), where he works with Prof. Daniela Rus. He developed robust and trustworthy machine learning models with particular breakthroughs in interpretable, safe, and verifiable AI.
His work has been recognized and published at top-tier venues, including flagship conferences like NeurIPS, ICML, ICLR, and AAAI. Furthermore, his works have also been featured in the high-impact journals Nature Machine Intelligence and Science Robotics, reflecting his contributions' significance and impact.
Mathias completed his Doctor of Philosophy degree (PhD) in 2022 at the Institute of Science and Technology Austria (ISTA) under the able supervision of Tom Henzinger. Prior to his PhD, he received this distinguished MS and BS degree at the Vienna University of Technology (TU Wien). This talk was given at a TEDx event using the TED conference format but independently organized by a local community. Learn more at https://www.ted.com/tedx