Learning Fast and Slow: Levels of Learning in General Autonomous Intelligent Agents