It's better to say that unsupervised ML is merely indirectly supervised ML.
Wasn't there a second version of AlphaGo that was trained only by playing against itself without external input and turned out to be even better than the first version that was trained with external input?
Isn't that unsupervised learning?
You have ground truth on the result: one agent won the game.
With voice recognition, that's a lot more vague and human supervision is helpful.
the rules of the game, and the output from winning, is supervision.