On the contrary, it was almost certainly computed using supervised training. Some set of people must have selected and labelled the training data. Their biases are cooked directly into the resulting software.