Some AI researchers believe that using learning methods with no built-in prior knowledge and throwing a bunch of compute at them is the path to building effecti

Some AI researchers believe that using learning methods with no built-in prior knowledge and throwing a bunch of compute at them is the path to building effective AI. I'm thinking of Richard Sutton in particular: - Bitter Lesson essay: http://www.incompleteideas.net/IncIdeas/BitterLesson.html http://www.incompleteideas.net/IncIdeas/BitterLesson.html - A lecture of his on temporal difference learning, which is a "model-free" method of reinforcement learning: https://www.youtube.com/watch?v=LyCpuLikLyQ https://www.youtube.com/watch?v=LyCpuLikLyQ I personally don't agree with his emphasis on model-free learning, but it's not the case that people are building model-free RL agents because they don't understand the trade off that they're making.