From: Ben Goertzel (ben@goertzel.org)
Date: Thu Feb 07 2002 - 22:02:45 MST
> If the goal is to instigate
> a Singularity as soon as possible, the right course
> of action may not be to start coding right away.
> It might be to work on AI architecture and
> algorithms, because clever solutions of such will
> allow you to use smaller hardware, and thus
> afford the required hardware sooner.
It's a good general point. My practical intuition is
that a roughly equal mix of implementational experimentation
and conceptual/mathematical theory is appropriate, at
this stage.
The hardware/software we have is good enough that we
can learn things about real AI from experimenting with
software, even without a full mathematical design.
On the other hand, it's a mistake to try to engineer
a full seed AI without a full mathematical design in
hand first.
The task of engineering a seed AI has all the problems of
large-scale software engineering combined with the problems
of cognitive science theory! It's quite a handful...
At Webmind Inc. we started out with a full mathematical
design, but our experiments caused us to revise it, and the
intermixture of conceptual design revisions and software revisions
became quite intense.
With my new project, Novamente, things are simpler because
we have a smaller team, and because we now have more experience
so the conceptual and software revisions that come up
are much smaller.
It is definitely a big mistake to ignore boring, practical
lessons about system engineering when trying to build a seed
AI. It may be a philosophically fascinating software system,
but it's still a software system, and routine software
engineering issues can fuck you up just as surely as
philosophical errors ... the combination of the two problems being
the most, er, interesting kind of phenomenon one
encounters...
-- Ben G
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT