Re: [agi] What best evidence for fast AI?

From: Joshua Fox (
Date: Sun Nov 11 2007 - 13:22:25 MST

> OK then, what are the arguments suggesting a single insight has a good
chance to
> make this much difference?

I don't know about a single insight, but I'd venture that a solid theory of
intelligence is the major obstacle between us and AGI.

If engineering an AGI does prove to be difficult, we can hope that the
theory will be clear enough and make enough of an impact to get all needed
financial resources.

An AGI, as Eliezer has said, does not have to imitate the details of the
brain, but it will broadly imitate its functionality. Airplanes easily beat
out birds in some characteristics once the theory was in place, and after
the unveiling of the Wright brothers' design it took only a few years until
planes were good enough to use in war. (I understand that controlling the
craft's balance was the key insight.)

Creating an intelligence is unlike building a bridge, building, or tunnel;
getting ships across the ocean; getting a rocket into space; or building the
atom bomb, which require vast resources of materials and energy even when
the theory of how to build one is understood. The human brain, the
existence-proof of intelligence, is just a one-kilo lump.

Although we do not understand the brain perfectly yet, it is becoming clear
that there is no super-engineering-trick there that threatens to transcend
the abilities of human engineering. Indeed, brains range from incrementally
from worms to human, and if we can engineer something better than the worms,
we can work our way up to human.

All the above doesn't set a time for AGI, but it does suggest that it might
come soon after the theoretical breakthrough.


This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:00 MDT