This would require a scenario a lot like in the podcast we were talking about, where there's a government-led project to get to transformative AI, and then rather than using that AI to dramatically help all humanity, the government instead decides to ban using AI to dramatically help all humanity (as a side effect of affirming the status quo and banning all uses of AI that threaten its own power), while still allowing limited access to this AI technology by the wealthy and powerful.
I actually don't think this is that likely, despite the fact that some people claim to be aiming for this future (or some similar future where humans remain in control and capitalism doesn't suffer a discontinuity). Even assuming this AI project doesn't kill everyone or otherwise go wrong, I think in an egalitarian setting there's overwhelming pressure to take transformative actions (save peoples' lives etc), and even in a dictatorial or plutocratic setting there's a lot of pressure to take transformative dictatorial actions (for your basic hedonist: kill off everyone they don't care for to save resources, for your more refined dictator: subtly arrange events so that their preferred political decisions work wonderfully and produce a flourishing civilization full of people who view them as a great leader).
(Edited because my previous reply was a bit off the mark.)
I don't think this scenario depends on government. If AI is better at all jobs and can make more efficient use of all resources, "AI does all jobs and uses all resources" is the efficient market outcome. All that's needed is that companies align their AIs to the company's money interest, and people use and adapt AI in the pursuit of money interest. Which is what's happening now.
A single AI taking dramatic transformative action seems less likely to me, because it'll have to take place in a world already planted thick with AI and near-AI following money interests.
Tangent to this post, but I read it by listening to the narration, and there are substantial differences between what the narration says is the text of this post and what text actually appears on the screen. I've noticed a smaller version of this with other posts in the past, but this time it seemed especially notable.
We work with T3Audio on the narration, and I think they don't really update it after initial publication. It costs us some non-trivial amount of money (like $1 or so) to narrate a post, which means we can't just re-narrate it on every edit without opening ourselves up to burning a bunch of money without reason. Not sure what the ideal thing here is.
Maybe instead narrating posts automatically when published, the poster could be shown a message like “Do you want to narrate this post right now? Once narrated, the audio cannot be changed.” And if they say no then there’s a button they can press to narrate it later (e.g. after editing). And maybe you could charge $1 if people want to change the audio after accepting their one free narration?
Historically, a purposeful decision that permanently lifts the whole population out of poverty was never on the table. Overall indifference doesn't prevent occasional philanthropy, but philanthropists were not that rich. So if there is some alignment (in the pseudokindness sense), the main issue is surviving until some group that cares gets rich enough. Which is not straightforward, since destruction of biosphere is a default side effect of post-human scaling of industry, and moderation in overall indifference about humanity is crucial at that step.
I think hoping for "pseudokindness" doesn't really work. You can have one-millionth care about a flower, but you'll still pave it over if you have more-than-one-millionth desire for a parking lot there. And if we're counting on AIs to have certain drives in tiny amounts, we shouldn't just talk about kindness, but also for example desire for justice (leading to punishment and s-risk). So putting our hopes on these one-millionths feels really risky.
Pseudokindness is not quite kindness, it's granting resources for some form of autonomous development with surviving boundaries. The hypothesis is that this is a naturally meaningful thing, not something that gets arbitrarily distorted by path-dependence of AI values, that is path-dependence mostly reduces its weight, but doesn't change the target. Astronomical wealth then enables enclaves of philanthropically supported descendants of humanity, even if most AIs mostly don't care.
The argument doesn't say that there aren't also hells, though on the hypothesis of naturality of pseudokindness that would be a concurrent thing, not an alternative. I don't see as strong an argument for their naturality as that for pseudokindness, as this requires finding a place between not caring about humanity at all and the supposed attractor of caring about humanity correctly. The crux is whether that attractor is a real thing, possibly to a large degree due to the initial state of AIs as trained on humanity's data.
Thanks for this! I ended up reading The Quincunx based on this review and really enjoyed it.
As an aside, I want to recommend a physical book instead of the Kindle version, for a couple reasons:
(Also, without the physical book, I didn't realize how long The Quincunx is.)
Even with those difficulties, a great read.