The political debate is endless and not very enlightening. However, this stuff max constitute useful tools to make better models in a non political way. A model that is biased because of too narrow test data will perform worse than a model trained on a suitable dataset. It can be hard to spot these issues, and having some checks and bounds to avoid common sources of bias is a good thing. Furthermore, an analysis might expose more info about the users than intended. This can have pretty bad consequences. For this purpose I thought the one about "collecting sensitive information" was a pretty good start.