Designing societally beneficial Reinforcement Learning (RL) systems