Reinforcement Learning
What an agent receives after performing an action
Reward
One limitation of AlphaGo
Can't play anything besides Go/can't transfer skills from Go to a different game
What did a group of researchers reveal about Deepmind's understanding of Breakout?
That it had no understanding of the game beyond the pixels (it didn't understand what a paddle was)
The ability of a program to transfer what it has learned about one task to help it perform a different, related task
Transfer learning
What are Deep Q-learning systems still failing at/unable to achieve
abstraction/domain generalization/transfer learning
What common problem does AlphaGo's deep Q-learning method's inability to be used to learn other tasks illustrate?
The "easy things are hard" paradox
The game platform used by the DeepMind programmers in their work on reinforcement learning.
Atari
Mitchell notes that while this program is impressive, it is far from any general AI application
AlphaGo
Mitchell claims that if an AI that could play this game better than a 6 year old child, that program would of "conquered several of the most challenging of domains for AI"?
Charades