This seminar will explore the challenges in designing and testing safe AI-based systems. Each week we will have a speaker from industry or academia discuss one problem area in AI safety. Our speakers will discuss issues such as verification and validation of AI systems, robustness, reward misspecification and hacking, explainbabililty of black-box models, AI ethics, and AI governence. Application areas will include transportation, natural language processing, and medicine.


All but the first class on 3/30 will occur in person in Gates B1. We will take attendance using Google forms starting from the second week. If you are taking the class for credit, you need to attend (in person) 7 out of 9 talks (not including the first talk). Please record your attendance here and let the course staff know if you have a conflict and cannot attend in person.