Russell Wallace wrote:
I don't think this is an accurate paraphrase of Mike's statement. "X
is secret sauce" implies X to be _both necessary and sufficient_ (or
at least that the other ingredients are trivial compared to X) - a
type of claim AI has certainly seen plenty of. But Mike's claim, if I
understand it correctly, is that visual/spatial capability is
necessary but not sufficient for AGI. (A position I also hold.)
There is no clear distinction between perceptual and symbolic paradigms.
The more intelligent the AI gets, it would probably contain more
perceptual-deductive methods.
Suppose if you want to program a symbolic AI to move and rotate a
robotic arm. It would be difficult and computationally intensive if it
is purely implemented by discrete, atomic symbols. All the angles must
be expressed by each individual symbol.
The solution would be to use an abstraction, such as a floating-point
number to represent the angle of the arm. This improvement, however,
represents data continuously and uniformly like perception; contrary to
a discrete symbolic representation. This improvement is a step towards
perceptual processing.
So there is no clear distinction between symbolic and perceptual
representations. The more perceptual-like the system gets (the more
numeric fuzzy variables it has), the more efficient it would represent
some tasks (like angles).
Discrete symbolic representations are chosen to represent the
"important" characteristics continuous and ambiguous objects. Discrete
symbolic representations are *inductive* because they categorize only
the "common" attributes of a continuous representation. Symbolic
representations are not exhaustive and only represent a *subset* of the
"common characteristics" of continuous and ambiguous data.
Probabilistic symbolic reasoning is an improvement because it is more
perceptual-like as it represents data numerically like perception. As in
the robotic arm example, increasing the angle would be easy if the angle
is represented numerically, instead of convoluted symbolic
representations representing every case and combination. The angle can
easily increment/decrement just like humans can move objects in their mind.
Thus, if the symbolic/associational AI stores some temporary
non-symbolic variables that it could modify, the variables are analogous
to perceptual reasoning because they can be easily imagined/modified
just like human imagining and modifying their visual perception.
If you make a symbolic AGI to use mathematics to represent
three-dimensional data, it is *already* a perceptual-like representation
because the mathematical variables or coordinates can be easily
incremented/decremented in a domain-specific way just like humans
imagining objects.
-------------------------------------------
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
