A change in scale is a change in kind. There's a real qualitative difference between a party popper and a flashbang even though they're both "just" generators of light and sound. It's the same with regular ol' human learning---for which we have millenia of understanding and norms---and comparatively low-cost global-scale machine learning. The scale of the latter obliterates the structures we've built under the assumption of the former, so you can't just claim they're the same.