On May 7, 2006, at 11:09 AM, Eugen Leitl wrote:
On Sun, May 07, 2006 at 09:29:51AM -0400, Ben Goertzel wrote:
If you explain to me what non-logic based reprsentation you prefer to
use to represent a particular piece of knowledge, I can almost surely
What if you don't yet know which piece of knowledge you have to
represent? And that there are no clear specs, and the requirements
change constantly?
There are levels of abstraction here; the nice thing about silicon is
that it lets you do almost anything, granted with wildly varying
efficiency depending on the problem. On the surface, first-order
logic and NAND gates are obscenely inefficient at dealing with the
situation you describe but there is no particular reason one could
not implement a more efficient abstraction on top of them.
Evaluation under uncertainty is a reasonably well understood
theoretical domain even if engineering practice does not exploit it
well.
Do you know any AI rewriting the code base (whether at SEXPR or FPGA
gate level)?
There is another way using somewhat different assumptions. A proper
non-axiomatic computational system is constantly updating the model
that determines the results of expression evaluation. Adaptive (and
particularly deeply Bayesian) expression evaluation in something that
resembles a functional language is essentially no different than
rewriting the code base even if the tiny core that is the virtual
machine a person wrote does not get rewritten. Noisy signal,
incomplete data, and errors are handled robustly; that is most of the
point of using non-axiomatic models in the first place. I would
guess that Pei Wang could add something to this.
Classic lambda calculus functional languages (LISP/Scheme being
canonical) are not well suited for non-axiomatic adaptation (IMO&E),
but there are other more obscure functional models (e.g. the
functional FORTH-like research/academic languages) can be made to fit
much more neatly though there are none that I would point to as being
ideal. It is an obscure corner of computer science even without
trying to integrate non-axiomatic concepts into it, so it is not the
sexiest of discussions. I've been meaning to drop this into a thread
for a year or so, but never got around to it; I've been using a non-
axiomatic machine model for a while. A lot of this is arguably dull
slog material -- tedious, unsexy, and useful.
None of which prevents memory-attached FPGAs from being very useful
for the broader task. I look forward to evaluating such systems.
Cheers,
J. Andrew Rogers
-------
To unsubscribe, change your address, or temporarily deactivate your subscription,
please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
