Re: 10x?! (was guaranteeing Friendlyness)

From: Ben Goertzel (
Date: Mon Dec 05 2005 - 11:45:47 MST

We are not initially focusing on programming a programmer. I think
that is the wrong place to start.

Our initial focus is on programming a baby, and then bringing up the
baby through Piaget's stages of cognitive development. Programming a
programmer will come ~5 years after funding.

Right now we have our core AI architecture implemented (in a single
machine version, extendable not too trickily to distributed computing)
and initial versions of our core learning and reasoning algorithms in
place. We are working toward having an artificial baby "living" in a
3D simulation world, but progress is dragging because nearly all of us
on the team have a lot of other responsibilities too. We hoped to
have the artificial baby learning some Piagetan baby stuff by the end
of 2005 but now it looks like it will happen sometime in early 2006.

What we do have is a fairly detailed mathematical and conceptual and
software design (embodied in a 1000+ page design document) that
explains how one can get from the artificial baby all the way through
to the artificial programmer. But there are plenty of details to be
worked through in implementing and refining the design, it's not a
design that's detailed the level where we could simply give it to
non-AI-expert programmers to code up.

-- Ben G

On 12/5/05, Herb Martin <> wrote:
> > From: Ben Goertzel
> <snip>
> > I am almost certain that a human-level self-modifying AGI could be
> > created with a few thousand machines right now, and I think it might
> > be possible to get by with a few hundred if one paid enough attention
> > to optimization. Based on these (admittedly rough) estimates, a
> > tenfold increase in computer power would bring human-level AGI into
> > the domain of relatively easy affordability.
> >
> > -- Ben G
> What sort of progress have you already made on
> "programming a programmer"?
> Please tell us anything you are able (and willing)
> to offer....
> --
> Herb Martin

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:54 MDT