The Natural Abstraction Hypothesis: Implications and Evidence — AI Alignment Forum