[ExI] elon says this isn't cgi

[Mike Dougherty]

On Fri, Jan 1, 2021, 3:50 PM spike jones via extropy-chat <
extropy-chat at lists.extropy.org> wrote:

> Ja, your idea (I think) is to use a simulated Zoom meeting where the
> avatar is animated but looks like a 3D meat-world person?  Or did I
> misunderstand what you have in mind?
>
I actually think a Tweekie (you remember 80's Buck Rogers?) or BB8 (recent
Star Wars) would be task-specific hardware. For my preference, a Wall-E
with a Nintendo Switch for a chest would be sufficiently mobile to
automatically get itself into viewing position and present whatever image.


I wanted more of a pet style companion or wizard's familiar.  I can
understand needing a heavier duty machine to provide lifting assistance for
mobility. The problem with large humanoid robots is that they have
gorilla-strength and less understanding. Even if they're engineered to be
safer than the auto-autos, perception that T-1000 will kill us all is
difficult to overcome.

> Computer graphics hipsters, how close are we to having animation good
> enough to pass for a person on a Zoom window?
>
You clearly don't play enough xbox/playstation/computer games.

> >…The barrier to listen-bot (which may well be alexa in a teddy ruxpin) is
> that nobody will talk to it any more than they've ever talked to an
> inanimate object…
>
> I disagree sir.  Plenty of us poured out our hearts to Eliza back in the
> 70s, knowing full well we were talking to ourselves.  This would be better
> than Eliza in a way: it could call on encyclopedic knowledge of the world
> thru the internet and it could remember stuff we told it last time.
>
Who is "us"?  You and I are pretty happy with endless streams of text.
"Normal" people are more fickle.  Even Eliza bored me fairly quickly.
 That's why I like the human+AI hybrid... so people can nudge the algorithm
when necessary and the AI can cover for us when we're inattentive.

> >…I think the solution is that the device must be conversational…
>
> Ja.  A critical design feature is that it would remember what we told it
> last time.  Then we start telling the same story a second time, it could
> decide to listen again or have it ask questions, knowing where the
> discussion will go.
>
This is my main focus for AI: to distill the flood of detail into essential
facts. Unlike the unintelligible weights of a neural net, I'm hoping to
have higher lever level abstractions "fact trees" that humans can
manipulate (in VR) - or think of it like tieing a specific fly to fish in
the unfathomably deep knowledge lake, the fish that you reel-in (real-in?)
is the answer/reply to the question/query represented by that specific
fly.  (Wordplay is fun, innit?)

> >…This is where I know the GPT-n and some text to speach is probably "good
> enough" for a viable product.   However, if this project is an excuse to
> get started, the "getting to know you" bot could build a knowledgebase...
> and the telepresence operator interface could navigate those topics...
> imagine if the next caregiver can enter a room and continue conversation
> where the last caregiver left (with AI providing fill-in conversation in
> the interim)
>
> A lot to digest in that one paragraph.  Consider the terabytes of inane
> “conversation” available on teen chat sites.  Perhaps we could set up
> something that would somehow mine and catalog the terabytes of inane
> conversation on ExI and MENSA, then feed that back to the user.  That
> sounds kinda cool, and certainly a good reason to archive our inane
> conversations here: a data source for future chatbots.
>
Terabytes isn't as much as it used to be. It gets smaller every day.

>  Lotsa good stuff in your post Mike, thanks.  I am the scarecrow from
> Wizard of Oz: I would get on this and figure it out if I only had a brain.
>
How many rockets did you build entirely by yourself?  I think this project
plan requires a team right from the start.

>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20210101/998336c2/attachment.htm>