Of course, those labels aren't purely artificial. Once you define some measurement of similarity, you can identify clusters in a data set, if there are clusters

Of course, those labels aren't purely artificial. Once you define some measurement of similarity, you can identify clusters in a data set, if there are clusters to be found. Those clusters are reasonable things to lump together with labels.