pnrjulius comments on Guardians of the Gene Pool - Less Wrong

25 Post author: Eliezer_Yudkowsky 16 December 2007 08:08PM

You are viewing a comment permalink. View the original post to see all comments and the full post content.

Comments (73)

Sort By: Old

You are viewing a single comment's thread. Show more comments above.

Comment author: stcredzero 09 June 2012 03:07:38PM 0 points [-]

AIs are designed, so they're optimized for whatever you optimize them for.

My prediction: The ones optimized for reproduction are the ones that will be around in the long term.

Comment author: pnrjulius 11 June 2012 01:33:51AM 2 points [-]

Not necessarily, because there's no law saying that AIs have to die. This changes the evolutionary calculus significantly; you don't need to reproduce if you can just keep existing and expand your power over the cosmos.

But you're right, insofar as AIs that rapidly self-destruct and never reproduce are not going to stick around long. (I think this is actually a tautology, but it's a tautology with the character of a mathematical theorem---definitely true, but not obvious or trivial.)

It's also worth considering that there are different constraints between NIs and AIs though. NIs have to change gradually, piece by piece, gene by gene. AIs can be radically overhauled in a single generation. This gives them access to places on the fitness landscape that we could never reach---even places that are in fact evolutionarily stable once you get there.

Comment author: wedrifid 11 June 2012 01:46:12AM 0 points [-]

you don't need to reproduce if you can just keep existing and expand your power over the cosmos.

Apart from the practical lightspeed limitations. You do need to reproduce or in some other way split yourself into space-separated parts if you wish to expand your power over a sufficient distance.

Comment author: stcredzero 13 June 2012 04:48:20AM 0 points [-]

One of our mind children might read this someday and think, "Distance? What a quaint idea!"

Comment author: stcredzero 13 June 2012 04:47:02AM 0 points [-]

Not necessarily, because there's no law saying that AIs have to die. This changes the evolutionary calculus significantly; you don't need to reproduce if you can just keep existing and expand your power over the cosmos.

As wedrifid pointed out, that depends on what one can do about the lightspeed limit. And thermodynamics. I don't think not dying of old age changes evolution that much. Humans are prone to geriatric diseases because evolution can't do much for us past the reproductive years. Beings without a lifespan won't face that.

I highly doubt that no AI won't ever destroy another, though.

It's also worth considering that there are different constraints between NIs and AIs though. NIs have to change gradually, piece by piece, gene by gene. AIs can be radically overhauled in a single generation. This gives them access to places on the fitness landscape that we could never reach---even places that are in fact evolutionarily stable once you get there.

That just means that they'll evolve without the constraints of genetics, much as designs and memes do.

I think it's a mistake to treat superhuman AI as magic. In some contexts it will seem magic, but not all. Human habitations viewed from 10,000 meters look like growths of lichen. In some contexts, some dogs are "smarter" than some people. Human intelligence gives us a tremendous advantage over all other life on Earth, but it is not magic. Superhuman intelligence is not magic. It's just intelligence.