Connecting the good regulator theorem with semantics and symbol grounding — LessWrong