Stuart Armstrong | Why AIs Fail
Foresight Institute’s Intelligent Cooperation Group
Why AIs fail
Abstract: Current LLMs and other generative models display most impressive abilities along with some astounding examples of stupidity and inability to generalise. This talk will look into why this is, and whether we can expect these models to improve in the near future.
Bio: Prior to co-founding Aligned AI, Dr. Stuart Armstrong spent a decade at the Future of Humanity Institute at Oxford University doing deep analysis on the biggest threats confronting humanity including nuclear threats, pandemics, human extinction, space colonisation, and - above all else - AI. Focusing on the power and risk of AI long before others were aware of these issues, Stuart deepened our understanding of AI alignment and developed many novel methods of AI control, publishing extensively and accumulating thousands of citations. His theoretical work reached a stage where it became practical - and, he felt, imperative - to implement it. In 2022 he co-founded Aligned AI with the aim of ensuring that AIs remained safe and under human control. Continuing his trend from academia as a mathematician, he co-developed, with Rebecca Gorman, several advanced methods for AI alignment – ACE, EquitAI, ClassifAI – which have become IP, patents, and products for the company. Developing these successful methods has confirmed his core belief: that the challenge of getting AIs to follow human values not only must be solved, but can be solved, and will be solved.
He is also the author of Smarter Than Us, a mentor for the Foresight Institute, an Advisor of the AI Safety Camp, and has appeared in several documentaries ("Alien Worlds", "The Future of Life and Death", "Odyssey") and interviews on AI and the future of space exploration.
Foresight Institute’s Intelligent Cooperation Group
This seminar is part of Foresight's Intelligent Cooperation Seminar Series. To join future seminars in this program please apply here.
A group of scientists, engineers, and entrepreneurs in computer science, ML, cryptocommerce, and related fields who leverage those technologies to improve voluntary cooperation across humans, and ultimately AIs.
Nominate a seminar presenter/topic
Feel free to reach out to lydia@foresight.org with any questions.