AGI 2011: The Future of AGI Workshop Part 1 - Ethics of Advanced AGI

[sharpezor]

Thank you google for providing this. All hail the mighty and all knowing hand of google.

[tmtyler]

46:23 - Rewards in reinforcement learning are scalars - by definition. Reinforcement learning is *defined* as being learning driven by a scalar reward. If you *don't* have a scalar reward, it isn't reinforcement learning any more.

[aebelsky]

cool