Someone should maybe look into possible revolutionary socio-technological solutions to raising the sanity waterline?
Here's a cluster in concept-space:
I peer at this, and my sense is that there may be a latent "big idea" to be discovered here, a way to set up a social-media/large-scale-communications platform that (1) is naturally engineered to promote good epistemics or some other unprecedentedly positive social dynamics, (2) is highly profitable/useful, such that it would spread virally, allow to capture a lot of value by whoever sets it up, and thus be self-funding and self-scaling, in a way that rapidly leads to it having a large positive impact, (3) has the latter property intrinsically entangled with the former property, such that it wouldn't be more profitable to create an "evil" version of the same thing.
Previous literature: I believe @TsviBT floated a roughly similar idea here.
I don't have any specific suggestions here, unfortunately. The starting point would be to think deeply on the topic of what human epistemics/human communities/human belief systems "are", build some toy model of it, then consider what structures could be designed within/around that in order to cause desirable positive effects. For example, prediction markets reduce epistemics to an economic problem and then borrow the tools of mechanism design to solve it; social-media algorithms treat it as a black-box DL problem; this models beliefs as graph networks on which you can interfere, etc.
Also: note that the space of possibilities here is pretty broad, because nowadays you can design this system to be quite intelligent, by embedding LLMs or LLM-like systems into it. (Not just in obvious ways, either! See e. g. JEV's pitch of intelligent systems that are "built to make fast, structured decisions that software can use directly".)
Why would that be useful?:
Again, I don't have specific ideas at this time, but this is likely what I'd be working on if I weren't stuck dealing with this AI crap. If you don't have much hope for agent foundations/alignment or human intelligence amplification, but would like to search for a "moonshot" solution, this might be worth digging into.
(3) has the latter property intrinsically entangled with the former property, such that it wouldn't be more profitable to create an "evil" version of the same thing
Previous literature, kinda: https://www.lesswrong.com/posts/WyeYJ7cmeiSrXeCnM/you-can-t-use-it-without-becoming-like-me
I peer at this, and my sense is that there may be a latent "big idea" to be discovered here
I don't have this sense. Or rather, seems super duper difficult, out of proportion to how encouraging your examples seem (which is not to say hopeless). But if you can spend like 10 minute... (read more)