One more post of being argumentative, and then I'll try something more cooperative.
Jnyusa wrote:
And you can't define a singularity tautologically. You can't just say, well, if it didn't create an unbridgeable chasm then it wasn't a singularity.
But that is exactly the definition of singularity! No information can get across. None of these things being talked about are true singularities. Look, singularity comes from physics, right? A black hole? No information survives that. The mass survives, but that's it. The way it was constructed is lost.
To me the essential part of any kind of singularity is that what is on the other side is essentially unknowable.
Jnysua wrote:
Machines, whatever their self-learning capability, are made out of non-living resources. They don't regenerate naturally. They depreciate.
Living organic creatures depreciate as well. How do they get around this problem? Replication. The same might well hold for AI creatures in the future. What if they learn to recreate themselves out of raw materials?
Jnyusa wrote:
And the great, superlative advantage of mutation and natural selection is that they are not structured as binary systems. They rely, as near as we can figure, on randomness. And randomness is infinitely richer than binary choice.
But organic mutation and natural selection is built upon the foundation of DNA, which is, essentially, binary. I mean, you can write out any DNA sequence as a binary string easily, since there are only four nucleotides.
The randomness you're talking about comes in, I suppose, in imperfections in the replication of DNA. It's random what part of the DNA isn't copied properly. The new copy may lead to junk results most of the time, but it might, in some circumstances, lead to a beneficial change. You could certainly set up an AI system where the behavior of the *creature* is governed by a binary string ( like its DNA ) and then when it replicated random bits were switched around to simulate mutations.
Paradoxically, what randomness allows is endless failure. And for that to result ultimately in success, you need lots and lots of time. We could, probably, maybe?, program a non-regenerating physical system to self-teach and evolve chaotically, but neither depreciation nor human investment parameters could withstand the millions of trials and waiting time, with no outcome certain, required for a chaotic system to come up with a 'desirable' outcome.
You need lots and lots of *computing* time. Computers are getting faster all the time. Evolution can happen on a much faster pace in a computer than it can in the real world.
Statistics: Posted by Faramond — Thu Mar 03, 2016 3:59 pm