John, (01)
As usual I think you've misconstrued a few points: I don't "reject
logic", I don't think people "process vectors in their heads". (02)
All the same "knowledge soup" is a useful enough term. As far as I
know you are unique among ontologists for having identified it. I'm
happy to leave you to argue its significance. (03)
Good luck promoting your commercial analogy engine. As a form of
example-based reasoning analogy it is one response to the problem. It
misses the key complexity issue so will never be a complete solution,
but it has the quality of making decisions relative and so is a good
first step. (04)
You could say analogy enables you to make as many unique distinctions
as you have examples. You generalize one example to another. That is
better than systems based on fewer classes than examples. But does not
capture the power of generalization to make _more_ unique distinctions
than you have examples. (05)
(Of course you could base your analogies on sets of examples, and so
get something like my "vector" approach. I don't know if you have done
that yet.) (06)
Pat: I think your ears are still closed. You are not simply not
addressing the complexity issue I refer to repeatedly. (07)
-Rob (08)
On Fri, Oct 3, 2008 at 6:57 PM, John F. Sowa <sowa@xxxxxxxxxxx> wrote:
> Rob and Chris,
>
> Since both of you seem to approve of Selmer Bringsjord's paper,
> that is a good starting point for resolving this debate:
>
> http://kryten.mm.rpi.edu/connectionist_logicist_clash.pdf
>
> As Selmer says in his one-sentence abstract,
>
> SB> A careful adjudication of the connectionist-logicist clash in AI
> > and cognitive science seems to disclose that it is a mirage.
>
> The issue to be resolved is to analyze that mirage and show how the
> heat of this debate is bending the light to create the mirage.
>
> CM>> You have yet to identify a single problem. Can you articulate one?
>
> RF> Knowledge soup.
>
> Thank you for referring to the term I used in my paper,
>
> http://www.jfsowa.com/pubs/challenge.pdf
> The Challenge of Knowledge Soup
>
> But I used that term to characterize the immense complexity of what
> people have in their heads, most of which is derived from the immense
> complexity of the world, society, etc., as processed by the very large
> human brains. The challenge is more than just a single problem.
>
> My recommended way of addressing the challenge is compatible with
> Selmer B's recommendation of 'Ecumenical AI', which makes a judicious
> selection and integration of elements from both sides of the clash.
>
> As a basis for integration, I have recommended Peirce's semiotics,
> which provides a vocabulary that defuses the clash between symbolic
> vs. subsymbolic approaches by a simple strategy: recognize that both
> sides are talking about *signs* and that the first step is to analyze,
> classify, and relate signs. For a brief summary of that analysis and
> classification, see Section 2 of the following paper:
>
> http://www.jfsowa.com/pubs/pursuing.pdf
> Pursuing the Goal of Language Understanding
>
> The remainder of that paper discusses many issues, and the slides that
> accompany it may be easier to follow. In particular, see slides 26-37,
> which include screen shots of an application to language processing
> that uses a combination of both continuous math and symbolic methods:
>
> http://www.jfsowa.com/talks/pursuing.pdf
> Slides that accompany the above paper
>
> RF> Wasn't a reconciliation of the connectionist and logicist
> > paradigms the central theme of Simon Levy's presentation in
> > particular, of the references I gave?
>
> Yes, he was trying to do something along the lines of processing
> low-level (subsymbolic) signs to support logic (symbolic signs).
> That is also compatible with Selmer's recommendation. But note
> that he didn't reject logic. He proposed new methods to support
> logic in a way that may be closer to what people do.
>
> RF> People commonly use vectors of contexts to represent word meaning.
>
> Just because some AI researchers use vectors to simulate some aspects
> of human cognition does *not* imply that people process vectors in
> their heads. That is just as much a non sequitur as assuming that
> people have theorem provers in their heads.
>
> RF> I built a system to break natural language sentences into
> > meaningful parts on this basis.
>
> That may be a useful mechanism to do one useful step in NLP, but
> there are infinitely more continuous methods than digital methods.
>
> I would encourage you or anybody else who has a promising method
> to develop it and demonstrate its potential. But finding the
> best (or even a good) simulation in an uncountable infinity of
> continuous methods is a nontrivial task.
>
> Good luck,
>
> John (09)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Subscribe/Config: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx (010)
|