Towards empathy in RL agents and beyond: Insights from cognitive science for AI Alignment — LessWrong