Anthropomorphic AI and Sandboxed Virtual Universes — LessWrong