#celestai

If you've never read FiO, you can find it here.

Alright, this post is gonna be huge. (starting foundational assumption: AI or future solid-state civilization that spreads out over the universe and converts it into computronium, kind of like CelestAI. If this assumption seems impossible, quit reading now.) So, this is about my best guess as to how the thing above would play out. Starting assumptions: No way to harness neutrinos for energy. No way to send any replicating structures into stars, neutron stars, white dwarfs, or far into planet interiors (too hot, too dense). No space-time engineering or femotech. No way to break light speed. Nanotech exists, but it can’t break the laws of physics (converting a cubic meter of rock to a cubic meter of computronium is gonna take a whole hell of a lot of energy for melting and shit). Technology to create mini black holes (around asteroid mass) exists. Reversible computing exists. Von Neumann probes exist. Efficient (let’s say 50%) light-energy converting nantenna panels exist. Dark matter is in WIMP form, and can interact and annihilate with itself, and gets cleaned from the galaxy by 10^20 years. (Actually a very pessimistic assumption). Goal is to maximize amount of computations done. The pathway taken is simple in retrospect. If we have an extractable mass-energy budget from a volume of space, you can get more computations done with more energy. So considerations favor most of the available matter going towards powering the computronium, and a relatively tiny portion going towards the actual computronium. (though of course you need matter to build energy collectors and stuff like that). Also, due to the landaur limit (it takes a minimum of boltzmanns constant*temperature*ln(2) joules to erase a bit), your computing nodes should be as cold as possible. So, waiting for the ambient temperature to drop from 3 K to 0.3 K ramps up your computation capacity tenfold. Would 90% of the mass-energy get lost in however many trillions of years that takes? No. So there’s an incentive to wait for a really really long time until the universe is cold and dead and the last star has burned out, before starting the party. One final consideration: speed. You could go faster. This would mean you would be spending more energy on launching spaceships (practically, more like small replicating probes). But you would also be getting there faster. You arrive one second faster to a galaxy, that turns one galaxy-second of power output towards your glorious civilization. And there are some galaxies that are far enough away that you’d need to go close to light speed to even reach before they vanish beyond the cosmological horizon. So with upsides that big, there’s overwhelming incentives to get as close as physically possible to lightspeed. Gotta go fast. Really fast. Like >0.9 c fast. Also, stars and planets are inconveniently leaking loads of energy into the void, gotta stop that. So, the “AI eats the universe” scenario would look something like “Build dyson swarms to keep stars from leaking out all this damn energy. Use energy to starlift so the star doesn’t go kaboom inconveniently. Store excess energy. Launch starships at ludicrous speed to everywhere in the universe. Wait. Wait. Wait some more. (hundreds of quintillions of years later when the universe is cold and dead) Use almost all this stored-up mass and energy to run glorious post-singularity civilization (though you’d never notice anything from the outside).” Spread out across the universe, turn off all the stars, wait, slow-burn the universe for fuel deep in the degenerate era of the heat death of the universe. Calling Anders Sandberg, @antisquark

Sisterly big butt bump.