Re: AI:This is how we do it

From: Eliezer S. Yudkowsky (
Date: Thu Feb 14 2002 - 07:15:25 MST

Eugene Leitl wrote:
> On Wed, 13 Feb 2002, Zero Powers wrote:
> > While the networkable OS may not be conscious, who cares? As long as it
> > can perform as well (or better?) than the sentient Jupiter Brain some
> > here are expecting.
> The network will not awaken, unless it is engineered to do so. Given the
> sorry state of IT security the perversion potential is however
> quantitative. With 1 GBit Ethernet being deployed and 10 GBit Ethernet
> being designed the local bandwidth is slowly moving into interesting
> orders of magnitude.

To be precise, it requires an enormously higher amount of computing power for
the network to awaken accidentally than for us to waken it deliberately, and
latencies and small node sizes raise the bar even further. If all the
Internet were interconnected over OC3 pipelines and nobody was doing anything
else on the nodes, "the Internet" still wouldn't awaken spontaneously until
well after I would expect AI to be deliberately achieved on some local
supercomputer. (Interconnecting the Internet and then deliberately trying to
create AI is a different matter, of course.) The point is that spontaneity
just doesn't work. Spontaneous generation of AI on the Internet is like unto
the spontaneous generation of mice from straw and dirty shirts, as I believe
Pat Cadigan once observed. If you have a planet-sized mass of straw and dirty
shirts and you wait 850 million years it will happen, but genetically
engineering mice would be a lot faster.

Eugene Leitl is right about the potential for perversion though.

-- -- -- -- --
Eliezer S. Yudkowsky
Research Fellow, Singularity Institute for Artificial Intelligence

This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 13:37:39 MST