RE: AI in <what?>

From: Justin Corwin (thesweetestdream@hotmail.com)
Date: Sun May 19 2002 - 17:16:13 MDT


Thanks for the reply, Ben, some thoughts below...

>From: "Ben Goertzel" <ben@goertzel.org>
>Visio-spatial organization is useful but also limiting, and it is only one
>among very many possible ways to structure thoughts.
>I think Novamente *should* be given camera eyes someday. However, I don't
>think this is *necessary*, just a convenience.
>Indeed, the directory structure is limited, more so than the visuospatial
>world. What if an AI modeled part of its concept hierarchy on structures
>it saw in chem and bio databases? A lot of interesting metaphors there....
>How about weather satellite data from around the world? There is a lot
>more richness of data on the Net than Web pages and directories.
>
>I fear that you're anthropomorphizing to a large degree, assuming that the
>human perceptual-manipulative environment is somehow the best one...
>
>Maybe to an AI that grew up on the net, the spatially-limited perspective
>of
>an individual human would seem terrifyingly, painfully limited... even
>"wonky" as you say ;.>

Well, quite so. My worry is not that human spatial perception and
organization is better/best. It's that the net kind of sucks. Now, dont' get
me wrong, it's possible that a mindsystem with it as it's native environment
might see emergent patterns and find this very enlightening and whatnot, but
the Net is largely a repository of knowledge about the physical world. And
without a way to relate this data, much of what it finds is going to be
useless.

So I worry.

> > NOW. Just what level of complexity in environment is needed for
> > AI to make
> > it all the way to AGI?
>
>I don't think we have any theory adequate to tell us that, it must be
>determined via experiment.

Of course we can't know that, I wasn't suggesting....
Okay, how about this. I was just wondering where the experiments should
start, that's all. I'd be perfectly willing to build Novamente a robot body
if I thought it was neccesary, but I'm not sure it is. But I think it can be
shown in theory what kinds of feedback and opportunities an AI should have.

>
> > But I have an intuition that mental organization is dependent on
> > environment
> > and what tools the AI has to interpret that. So an AI that lives
> > on the Net
> > might never work at all, because of insufficient environmental feedback.
>
>I agree, this is possible...

Heh. this doesn't sound like you're internally saying 'plausible, too' while
you write this.
My big worry is that the net, being largely made up of static images, won't
be able to give the AI enough to work with.

I'm drawn to the image Dossy made on Extropians, of his toddler playing with
a door to figure out how door hinges work. All the AI gets is pictures of
doors, except on local nets where it has permissions to play. That sounds
really sucky to me. How will it learn if it can't act?

>
>Or, as I suspect, it might work VERY DIFFERENTLY than humans because of
>VERY
>DIFFERENT environmental feedback!

heh. Well, yeah.

>I think that socialization requires more than a chat window; it requires
>interaction with other minds in a shared perceivable/manipulable
>environment

Well. I wasn't saying it HAD to be a chat window, I was just saying that
chat windows are an example of non-physical social interaction. Novamentes
transferring direct mindstuff, and Eliezertrons passing flarespeke are both
examples of non-physical socialization.

>
>So I think that it is pretty much on a par with the other things you
>mention

Oh, I think it could be MORE important than the points I brought up. I just
dont' see it as an aspect of explicit environmental design.

>
>Not a critical point though!
>
>ben
>

Now now ben, with the error tolerances we're dealing with, they all could be
critical points!

Justin Corwin

"I like the word 'sucky', it has both emotional and realistic associations
for me."
              ~Dennis Leary

_________________________________________________________________
MSN Photos is the easiest way to share and print your photos:
http://photos.msn.com/support/worldwide.aspx



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:38 MDT