Rich Sutton on Reinforcement Learning paths to AI: Alpha Go Zero to 60
A fireside chat with Rich Sutton from Google DeepMind and the University of Alberta, and the founding father of Reinforcement Learning: https://www.youtube.com/watch?v=QqLcniN2VAk&app=desktop
I started with an exploration of whether the AIs of the future will be human-centric or alien, and why? Why are goals central to intelligence?
He was very excited about Deep Mind’s new Alpha Go Zero results, whereby the training of the game-playing AI was dramatically improved by removing human training data sets altogether. It bootstrapped itself from just the rules of the game and a series of AI vs. AI matches. It took just 72 hours.
Summary: http://www.wired.co.uk/article/deepmind-alphago-zero-nature-reinforcement-learning
Video segment: https://youtu.be/QqLcniN2VAk?t=16m1s
Dr. Sutton is considered one of the founding fathers of modern computational reinforcement learning, having several significant contributions to the field, including temporal difference learning, policy gradient methods, the Dyna architecture. In June 2017, Demis Hassabis announced that Sutton would co-lead a new Alberta office of Deepmind, while maintaining his professorship at University of Alberta.
Leave a Reply