This group addresses the most pressing societal and safety challenges of the ongoing unprecedented development of Artificial Intelligence. We bring together public stakeholders and experts from various fields, such as AI researchers and machine learning engineers, tech entrepreneurs, philosophers, ethicists, cognitive scientists, biologists and many others, to discuss different aspects of AI risks and benefits and to promote the culture of safe and responsible AI development.
Since Nick Bostrom's book "Superintelligence", the AI safety research landscape developed quite a lot. And it is no longer true that only a handful of people are working on the problem.
We will overview some of the main research agendas currently pursued and describe some of the technical problems. Also, we will go through several possible framings of problems with advanced AI systems, which are different from the classical framing most known from Bostrom's Superintelligence and older writings of Eliezer Yudkowsky.
The talk will be somewhat technical, but should be accessible for anyone familiar with the arguments from Superintelligence and some knowledge of current ML techniques.
About the speaker
Jan Kulveit works at Future of Humanity Institute, Oxford University. His research is centered on studying the behaviour and interactions of boundedly rational agents and more generally on making AI aligned with human interests. He is also interested in modelling complex interacting systems, and strategies to influence the long-term future.
Previously he worked as a researcher at the Institute of Physics ASCR, was the Strategy Director for the Czech EA Association, and co-organizes the Summer school on Human-aligned AI. His background is in theoretical physics, phase transitions, networks and complex systems.