A brief review of the reasons multi-objective RL could be important in AI Safety Research — LessWrong