Models of AI Safety

Talk by Stuart Armstrong

March 29, 2:30–3:30 pm (Q&A to follow)
Arts 241

Talk Description: Ideas are easy to have. Ideas about something as vague as the risks and potentials of superintelligence–even easier. But the quality, truth and limitations of these ideas start to become apparent when you cast them into models. In this presentation, Stuart Armstrong will talk about the models of Artificial Intelligent behaviour, how to build them, how to refine and use them, when to discard them and what they tell us about dealing with AI risk in the decades to come.

Brief Bio: Stuart Armstrong received his PhD in mathematics in 2006 from the University of Oxford.  Afterwards, he was a Junior Research Fellow at the Erwin Schroedinger Institute in Vienna, where he specialized in geometry. Starting in 2008, he was a researcher at InhibOx Molecular Research, where he worked on computational biochemistry. Since 2011, he has been a James Martin Research Fellow and the Alexander Tamas Fellow in Artificial Intelligence and Machine Learning at Oxford's Future of Humanity Institute. He published Smarter than Us: the Rise of Machine Intelligence in 2014 and worked with the DeepMind collaboration on the topic of interruptibility, garnering attention in more than 100 media articles.

This is a joint Department of Mathematics & Statistics/Department of Computer Science colloquium.

Upcoming Events