carado.moe

tammy's blog about
AI alignment, utopia, anthropics, and more;

also on twitter, lesswrong, rss, and at

implementing the platonic realm

(or: holding the world by its souls; or: upside-down metaphysics)

i think, to construct a utopia (possibly especially a sublime one) we ought to reverse the metaphysics of the world.

in any graph forming a hierarchy, you can pick a different node as root and view the graph from there. for example, in the model of reality below, you can pick psychology as your root instead of physics, and end up with reversed arrows.

this gets closer to something like a "platonic realm" in which objects are truly, objectively integral, rather than patterns in the stuff they're made of. we can select any layer to be this, and then maybe let details fill in as we investigate them, de-implementing reductionism. we can implement this the same way video games implement objects: as their own data structures with a bunch of methods to manipulate them, some of which able to generate detail to fill the object with; but those details are not an essential part making up the object, it's moreso the object and its context that produce the details.

these top level platonic persons are still full persons, in the computational sense we care about: their full neural net is still being ran at the top level.

why do this? so we can make a world that is actually focused on, and aligned to serve, persons. to make persons legible to the superintelligent singleton running that world.

now, if other detail layers contain enough compute to simulate another person, then there might still exist suffering persons that the top level is not keeping track of.

two solutions are: either detect encoded persons and extract them back up to the platonic level, or make other layers too weak to consistently encode suffering moral patients — this would put severe bounds on what kinds of computers we can use, but maybe that's fine. maybe computers are not even worth having, in the person-focused platonic realm.