Re: Seed AI milestones - Explicit Knowlege Codification

From: ben goertzel (ben@goertzel.org)
Date: Thu Feb 28 2002 - 09:45:42 MST


> > 1) one had the right type of hardware to create a human-baby-like body
as
> > well as brain
>
> Actually, the brain is flexible enough to learn to drive a minimal
> telepresence puppet. Also, one would ideally build an artificial reality
> box, which apart from being exempt from physical constraints has the added
> advantage of being able to run much faster than realtime.

Sure, but a realistic artificial reality box requires more hardware than is
affordable at present... as well as a host of funky software innovations.
Fun stuff, but obviously a huge project in itself.

> > 2) one had a detailed map of the human brain and body to use as a
blueprint
> > for constructing a particular baby-like system using this type of
hardware
>
> We're certainly getting the imaging capabilities, especially considered
> destructive cryoAFM imaging of vitrified critters over the next two
> decades. Which is probable also the time scale at which usable hardware
> comes online.

We have different intuitions about what constitutes "usable hardware"
for real AI. I think that current hardware is about good enough, and the
problem
is having the right software implemented.

But, better hardware will certainly be appreciated.

> It would be really interesting to see this approach to succeed. I will be
> amazed, but I'll be the first one to open a bottle to celerbrate.

Well, if we do succeed it certainly will not have been easy. I've been on
this for 15 years and my collaborator Pei Wang has been working in parallel
on his own related ideas for just as long. Many many man-years of
prototyping
were done in the Webmind Inc. R&D group. I'm well aware that making this
approach work is hard, and I can appreciate the relative straightforwardness
of the brain emulation approach (not that there aren't dozens of extremely
hard
technical issues with brain emulation, but there is justifiable confidence
that
once these are solved success will be straightforward). But I still think
we'll
succeed ;>

ben



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT