Comparing AI to our other technologies to assess risk is flawed. AI will create technologies that we cannot; so we should not compare them to those we could.

Interesting take. 

(1) Would it not also have been applicable to computing generally? Or the internet? Or steam?
(2) If we cannot compare the future to the past, to what should we compare it?

General Counsel @valoraxyz | Casts are personal | I am not your lawyer 🕶️

No. AI will literally create technologies. It’s intelligence. We cannot compare it. Probably the best effort would be to a hypothetical more-intelligent species, but AI doesn’t have the limitations.

it starts as narrow tooling that can perform in a superhuman manner. fundamentally, it is incomplete and nonviable atm. we haven't demonstrated that it can be generalistically complete, but we can attempt to discover depth-first simulation of the worse ways it can propagate.

𝖒𝖆𝖓𝖉𝖆𝖙𝖚𝖒 𝖆𝖉 𝖆𝖘𝖙𝖗𝖆

what forces reality to imitate our imagination to completion? sure. there's kernels of similarity, but when it comes to modeling nondeterministic things like AI, how does this remain accurate for, say, 6 months out? are these training runs guaranteed?

I used to load Lemonade Stand on a PET computer from an audio cassette. It would have been difficult to imagine FC on this iPhone then. 🙂 I can easily imagine evolutions that make the current AIs and our ideas of them seem quaint.