Charles D Hixson wrote:
...I think the mistake here is presuming that intelligence is some
particular set of tools that can solve everything. It is my belief
that OTOH intelligence is a framework into which can be slotted a
(perhaps) almost infinite set of tools. Most of them will be special
purpose, some will be more general in capabilities. As just one
example people have some built-in hardware for recognizing faces. It
appears to be rather specialized, though it can be adapted to a few
other uses...but doing so often causes the thing being thus dealt with
to be anthropomorphized. Language is one of the more general tools.
So is an intuitive set of rules for judging trajectories. (This set
of tools is quite limited, but handles unpowered objects in flight
with remarkable efficiency.) And there are *LOTS* of other
specialized tools. None of them is intelligent, but they slot into a
framework that uses them. It's this framework that's
intelligent...though I will freely admit I don't understand exactly
what intelligent means. I'm rather sure it doesn't mean predicate
logic. That's one of the special tools (probably a sub-module of
language).
My general schema uses a four-way decomposition of the interface along
the lines of purpose, technique, costs/value, and model. (Sometimes I
use other descriptive words. They have to do with the data structures
that I believe are employed by the schema-face. As general terms I
use water, air, fire earth in homage to CGJung. (Water-purpose,
air-technique, fire-cost/value, earth-model). Of these only Water is
teleological. I normally call these goals...but others use that word
to mean very different things than I mean, so water is a better
general term.
NOTE: I'm *NOT* saying that these define intelligence. I'm saying
that this is the shape of the interface between intelligence and the
specialized tools...including language as one of the specialized tools.
I apologize for replying to my own post...but I left out the point:
Many of the tools are NP-hard to design. But the tools can be add-ons.
You don't need to wait until all the tools are built to build the
interface and the intelligence that deals with them. Unfortunately,
it's much less clear exactly what the intelligence does. It is the
coordination system between the relatively independent tools...but it's
techniques don't show on the surface. Generally it's only the
specialized tools that render themselves evident, because they handle
all, or nearly all, of the outwards facing operations. It's quite
plausible that the intelligence itself is much simpler than many of the
more complex tools that it uses. It's *possible* that designing the
intelligence is NP hard ... but I don't see any grounds for making that
claim. Making that claim for the tools, yes. Several of them do seem
to be NP hard...or rather special purpose solutions that are "generally
good enough" to deal with several problems that in their general form
are NP hard. Note, however, how we deal with trajectories. To catch a
ball flying through the air we do the equivalent of solving a *linear*
equation. We do this by moving our head a known distance to get a
snapshot to work with and then running in the indicated direction. Then
we take another snapshot and compare them. It's more complex, and I
don't remember all the details. But it doesn't handle the generalized
trajectory problem, only the version that our ancestors were most likely
to encounter. So frisbies generate trajectories that are quite
difficult to calculate, and so does a drop-ball pitch, or a curve ball.
We handle THOSE cases in the same way that we learn to play the piano.
I.e., we can't depend on the "special trajectory tool", but we can
combine it with the "learn repetitive actions" (cerebellum) tool in a
very effective way. And this STILL doesn't give us a general solution.
There's just a larger number of special cases that we can handle.
-------
To unsubscribe, change your address, or temporarily deactivate your subscription,
please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
