Hi,

>Somebody could write an excellent paper about the
> potential pitfalls of such an approach (detail, fidelity, deep causality
> issues behind appearance, function, and inter-object + inter-feature
> relationships, and so on).  If nobody else is working in detail on
> publishing such an analysis perhaps I will study those issues for some
> months and try to write something for AGI-09 about it.

If you'd like to collaborate on such a paper, I'd be interested.

I have thought for a while of writing a paper with a title or theme of

"What Must a World Be for an AGI to Develop In it?"

... the goal being to guesstimate a "requirements spec" for virtual worlds
for AGI ... or, more plausibly, different requirements specs according to
different AI paradigms

To me there are 3 relevant paradigms here

1)
Perception and action centric, which leads to the conclusion that
virtual worlds will only be good enough for AGI when they're very
close to the richness of the real physical world

2)
Socialization, logic and language centric, which leads to the conclusion
that current virtual worlds are likely good enough

3)
Integrative, which leads to the conclusion that current virtual
worlds probably are NOT good enough, but that some relatively
moderate improvements to them could make them so

The sorts of improvements I'm thinking of are stuff like

-- replace animation based skeleton control with Player/Gazebo
style servomechanism based control

-- make more objects decomposed of parts in a meaningful way ...
including enabling stuff like slicing a cake in various ways with a
knife, ripping a page out of a book, etc.

-- give the AI more detailed feedback regarding interaction of
parts of its body w/ the external world, and also some internal
body feelings related to things in the world

-- simplified, not necessarily realistic fluid mechanics
[this one is a nicety, not a necessity, but it would certainly
help with understanding a lot of NLP metaphors] ... having
a world consisting of things only in one state of matter is
somewhat conceptually limiting...


-- Ben

-------------------------------------------
agi
Archives: http://www.listbox.com/member/archive/303/=now
RSS Feed: http://www.listbox.com/member/archive/rss/303/
Modify Your Subscription: 
http://www.listbox.com/member/?member_id=8660244&id_secret=101455710-f059c4
Powered by Listbox: http://www.listbox.com

Reply via email to