I wrote my recent Accelerando post to mostly stand on it's own as a takeoff scenario. But, the reason it's on my mind is that, if I imagine being very optimistic about how a smooth AI takeoff goes, but where an early step wasn't "fully solve the unbounded alignment problem, and then end up with extremely robust safeguards[1]"...

...then my current guess is that Reasonably Nice Smooth Takeoff still results in all or at least most biological humans dying (or, "dying out", or at best, ambiguously-consensually-uploaded), like, 10-80 years later.

Slightly more specific about the assumptions I'm trying to inhabit here:

1. It's politically intractable to get a global halt or globally controlled takeoff.
2. Superintelligence is moderately likely to be somewhat nice.
3. We'll get to run lots of experiments on near-human-AI that will be reasonably informative about how things will generalize to the somewhat-superhuman-level.
4. We get to ramp up [...]

---

Outline:

(03:50) There is no safe muddling through without perfect safeguards

(06:24) i. Factorio

(06:27) (or: Its really hard to not just take peoples stuff, when they move as slowly as plants)

(10:15) Fictional vs Real Evidence

(11:35) Decades. Or: thousands of years of subjective time, evolution, and civilizational change.

(12:23) This is the Dream Time

(14:33) Is the resulting posthuman population morally valuable?

(16:51) The Hanson Counterpoint: So youre against ever changing?

(19:04) Cant superintelligent AIs/uploads coordinate to avoid this?

(21:18) How Confident Am I?

The original text contained 4 footnotes which were omitted from this narration.

---

First published:
October 2nd, 2025

Source:
https://www.lesswrong.com/posts/v4rsqTxHqXp5tTwZh/nice-ish-smooth-takeoff-with-imperfect-safeguards-probably

---



Narrated by TYPE III AUDIO.