Alignment As A Multi Agent Intrinsic Reward