This doesn't assume that data is of a certain quality. In fact, many people that analyze data assume data is broken, which is why any analysis work, ML based or other, needs to spend more of its time cleaning and trying to work through problematic data. Only those companies that are very naive or starting out think otherwise.

There are other problems when trying to generalize too often from a large data set and apply that to individuals. Again, that's often a misapplication of data to a problem, which is either too general or ill defined.

It can be scary or problematic for any one large scale collection of data, but that's entirely separate from the potential to actually innovate new applications using the data.