A proposed system of ethics (for instance: consequentialism with target = Diversity) must be evaluated in some fashion in order to determine its value. However, any system of evaluation must make value judgements, and how do you do this without first fixing some “higher value” system? This seems like an unsupervised training problem, we don’t have labels for good or bad prior to fixing a system of ethics. Here are some potential principles for completing such a task.
The proposed system should reproduce core beliefs about right and wrong. If your proposed system of morality never condemns murder, it probably shouldn’t pass the sniff test.
It should provide some sort of insight beyond prescribing default human activity If your proposed system of morality doesn’t do anything else besides prescribe generally nice, kind, agreeable behavior, it probably sucks.