[CamperBob2]

The history of modern ML is just fascinating, and as far as I can tell it's utterly unprecedented.

    1945-2012   Let's figure out how we can build smart machines
    2012-2017   Wait, what the hell... we just needed more gates?
    2017-?      Let's figure out how this machine we built actually works

[swatcoder]

Far from unprecedented, just radically different than what people are used to in computing.

There are a lot other domains whose history (and present) on semi-informed stumbling into something effective and then spending decades (or lifetimes) trying to reverse engineer how it works, when it doesn't work, what the peripheral consequences are, and how impactful those consequences are.

Metallurgy and material science, agriculture, chemistry, pharmaceuticals, psychology, etc etc

20th century discrete/digital computing and computer science, having hewn close to mathematics and logic for most of its life, is actually the more unprecedented history as far as practical sciences go.

The flipside of all this is that all those other practical sciences have come with really very messy histories in terms of unintended consequences and premature applications, and (for better or worse) we can anticipate the same here.

[CamperBob2]

Great points re: pharmacology and psychology, certainly. I was thinking more in terms of technological applications. Normally the science comes first, followed by the tech, but AI has flipped the paradigm.