But to me this is not a Pareto improvement (i.e., this is subjective). I consider the bolstering of the hold Google (and Apple, and Amazon, and Microsoft) has over a number of common domains (and in a broader sense our lives) to be undesirable and unethical. So when I correctly teach Google's proprietary image recognition software what a traffic sign, mountain, store front (ugh, these are the worst), or tree is, I am (even in the strict economic sense of a Pareto improvement) harming myself.
Going from throwing away the work to improving Google's image recognition is a Pareto improvement so I'm interested in the justification.