The goal is absolutely logical (training the model with data that is undesirable). It is the method you are questioning? I guess I would have just labeled them

The goal is absolutely logical (training the model with data that is undesirable). It is the method you are questioning? I guess I would have just labeled them "terrible" or something in a bulk assignment to not have people observe them.