[extropy-chat] AI design

Alejandro Dubrovsky alito at organicrobot.com
Thu Jun 3 15:48:04 UTC 2004


On Thu, 2004-06-03 at 16:52 +0200, Eugen Leitl wrote:

> On Fri, Jun 04, 2004 at 12:41:07AM +1000, Alejandro Dubrovsky wrote:
> 
> > It's also a pity that you dismiss all other projects as useless. >H
> > non-feedbacked  narrow(ish) AI (eg Really Good theorem provers) could
> > make the development of a suitable RPOP much safer (or even at all
> 
> We don't have the crunch to move stuff in real world, nevermind software to
> write software rationally (i.e. not shuffling blocks, or banging around on
> instructions for peephole optimization a la ATLAS).

I meant them as an easier problem than going straight for the RPOP, not
as a weekend project.


> > obtainable).  Cutting deaths from 50+ million / year down to fuck all /
> > year also does not require such a huge jump in intelligence.  Yes, the
> > bar for a script kiddie to destroy the world is always lowering, but
> > that has been pretty low for a while in the biochem field and still here
> 
> There is no script kiddie culture in biosciences. Not even blackhats
> (military people are establishment). One has to own and manipulate equipment
> for that, or even to build it (explain that laminar flow table to Mom might
> be a tad hard).
> 
No script kiddie culture yet, no, but there's lots of grad and undergrad
people with access to the stuff, and none have dedicated themselves to
the task successfully.  On the same category in CS, script kiddies use
buffer overflows as almost their sole recourse (that i know of), a
technique probably first developed a couple of years before ENIAC.  None
use even elisa-level chats to fool people into revealing passwords since
it hasn't been packaged for them in the right way.  It'll take time for
a pluggable self-improving AI kit with 1337 haX0r instructions comes to
the net.

> > we stand.  (btw, an AI to deal with bio threats sounds more immediately
> > urgent and 10^9 times easier (i don't mean any uber-beast, just a
> 
> That's not a software problem. Docking and mutagenesis is useless, if you
> don't know what to dock, and what to mutate.
> 
> If you have millions of sensors beaming realtime structure/sequence data 
> from the field, then you've got something to work with. 
> 
No need for that level.  After the first couple of hundred dead, you get
samples, fix, distribute.

> > program that creates instant molecular solutions to viri and a suitable
> > distribution program))
> 
> What this planet needs, is a *working* machine learning in high-performance
> forcefield, using empirical and ab initio-ish constraints. This is something
> far easier than AI seed, but apparently not sexy enough for all the
> slashdotters.
> 
learning what?  or did i parse that wrongly?
alejandro





More information about the extropy-chat mailing list