Re: Seed AI (was: How hard a Singularity?)

From: Eugen Leitl (eugen@leitl.org)
Date: Tue Jun 25 2002 - 09:57:54 MDT


On Tue, 25 Jun 2002, James Higgins wrote:

> Eugen, are you an engineer? I believe Mr. Rogers was trying to say

No, just a science background.

> that generally referring to systems as "complex" has little meaning.
> At least unless you are using a rigid measure of complexity to
> quantify the implementation difficulty and time required. Just saying

While I was using the term complexity fuzzily, I can indeed quantify it.
Baby-shaped matter has a higher information content than the same atoms
after achieving chemical equilibrium.

> that something is complex does not meet this criteria, of course.
> Saying that life is complex and thus an AI must be complex to compete
> thus has no meaning in an engineering sense.

Luckily, I'm not an engineer. I'm free to say complexity without
automatically implying computational, and having man-month figures
floating at the back of my head.
 
> As he puts it an AI could easily be considered complex "in a
> pedestrian sense", which is true. But in an engineering sense no one
> yet knows exactly how complex a good AI implementation will be. The

I can give you an upper bound: a human baby. That's a ballpark, you can
go a bit lower.

How many bits do you need to describe a baby? (Remeber, not the zygote. A
a complete formed baby).

> complexity of the world, however, is completely irrelevant to the
> problem (since we're not trying to create an Artificial World, just a
> single Artificial Intelligence that would exist in it).

What I'm saying is that an AI has to have a large fraction of the world to
be represented in it before it even can start to learn. Because the world
is dirty, and complex, the resulting architecture is that, too. The vessel
may be elegant, but not the contents.



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:39 MDT