[ExI] Survival (was: elections again)

[Eugen Leitl]

On Wed, Jan 02, 2008 at 09:45:10AM -0500, Harvey Newstrom wrote:

> Please give more specifics than just this assertion.  What do you mean by take 

I can't believe you forgot about http://www.foresight.org/nano/Ecophagy.html
I don't subscribe to all its assumptions, but the whole scenario is realistic,
as in physically possible. 

> over the planet?  Convert to computronium?  With what?  With nanites?  How 

Taking over the planet, as in completely controlling energy and material
flows within the upper regolith. Blood music couldn't happen for physical reasons,
but this scenario is feasible. Vinge's Blight is another very realistic depiction,
minus the magic physics, of course. Most of Vinge's IT could be well be
patented already, and of some of it we already have prototypes.

> will it get a MNT factory built in a week?

Freitas cites minimum replication time of 100 seconds, which is IMO implausible.
(His energetic analysis might be unnecessary pessimistic, however, and he's
assuming a stealthy burn, since there's countermeasure around as well, which
is completely optimistic in the case of a transcendence scenario).
Otoh, we're talking about a system smart enough to bootstrap to the limits of
physics (not necessarily limited to what we think is possible, but when
estimating we necessarily must stick to what we know) very shortly, so it
could be as bad as that, or worse.
 
> Seriously, I keep hearing these predictions that once a recursive AI gets 

There's no such thing as a recursive AI. I can readily draw you up a 
recipe for a simple, orthogonal hardware that can run biologically inspired 
infoprocessing architecture at O(0) at 10^6 speedup at least. What currently 
prevents me from building prototypes of it is that such fabbing capacities are 
currently not available. I cannot order a mole of transistors, nevermind
a teraton of computronium.

> looping, it will transform the earth in a day (or a week or a month).  But I 

The duration is not important, what important is to know that a new
player is on the scene, and whether you can disrupt the bootstrap. 

> never get any specifics as to how, except that it's so much smarter than us 
> that the laws of physics don't apply to it.

Of course the laws of physics (not necessarily physics as we know it) do
apply.
 
> Is this whole prediction based on the assumption that the AI will do something 
> that we can't predict is even possible right now?  Does it require great 

No, it's just pulling the equivalent of http://www.caida.org/publications/papers/2003/sapphire/sapphire.html
on the physical layer, with the exception that you turn the entire planet's 
carbon sink into a zombie network. It can be really fast and hostile,
using everything mostly for fuel, or slow and constructive, but it would
be still TEOTWAWKI all over.

> unknown powers or exceptions to our current understanding of physics to exist 
> that we just haven't discovered yet?  How exactly is this "...then a miracle 
> occurs..." supposed to happen?

Imagine you've got a runaway worm that turns trees into routers.

-- 
Eugen* Leitl <a href="http://leitl.org">leitl</a> http://leitl.org
______________________________________________________________
ICBM: 48.07100, 11.36820 http://www.ativel.com http://postbiota.org
8B29F6BE: 099D 78BA 2FD3 B014 B08A  7779 75B0 2443 8B29 F6BE