Hi John, (01)
Thanks for the links, but that isn't what I am
interested in; I'm more focused on WHY Cyc (and
other approaches to developing massive knowledge
bases) is thought by so many to have failed, e.g.,
Genesereth for example. (02)
Before the Cyc project started, the common view
given full imprimaturitan status in the research
community was that it was KNOWLEDGE that was
lacking for full AI, and only THAT was keeping AI
from universal suffrage. (03)
I am looking for opinions by people who might know
as to WHY that assumption was clearly so wrong.
Knowledge is NOT enough, and that has been clearly
demonstrated by Cyc's lack of demonstrated value
in universality of intelligence. (04)
Small, highly focused projects, such as the blocks
world and its successful linguistic manipulation
as per Terry Winograd's (admitted) kluge, and the
surprisingly good results from very simple (also
kludged) chatbots such as Parry, and the
Somebody's Prize demonstrating lack of scalability
of said chatbots, shows SOMETHING. But what KIND
of something? (05)
More study of Cyc seems to belong to the same
viewpoint as theocratic studies of angel densities
and pinheads, the viscosity of ether, and mappings
of magnetic fields in thousands of points.
Instead, the Einsteinian approach of novel - dare
I say subjective -interpretations (in his case, of
the Michelson-Morley results) seems to be what is
most clearly lacking at this point in time. (06)
Why DON'T huge hunks of deduced, induced, abduced
and reduced knowledge suffice? What is still
lacking? Why don't gobs of special purpose
functionality, coupled with gobs of knowledge, do
the trick? (07)
Why DO simple approaches work so well at small
scales? (08)
Why DON't simple approaches scale well? (09)
Why DOESN't a simple chatbot with Cyc on its back
suffice to convince observers in a Turing test? (010)
In the fifties or so, game theory was developing.
Turing came up with a biological explanation of
what would be called the hox genes to form complex
biological strata. Lately, we have learned that
there are only some 20,000 genes which are
adequate for making a human bean, but that leaves
out a LOT of so called JUNK DNA, meaning genetic
structures we still don't have a clue about. (011)
Those are the kinds of new ideas that need to be
reduced to practice. And that is why the patent
form, with one advance teaching AGAINST prior art,
seems interesting to me as a model of how to take
the next steps. (012)
-Rich (013)
Sincerely,
Rich Cooper
EnglishLogicKernel.com
Rich AT EnglishLogicKernel DOT com
9 4 9 \ 5 2 5 - 5 7 1 2 (014)
-----Original Message-----
From: ontolog-forum-bounces@xxxxxxxxxxxxxxxx
[mailto:ontolog-forum-bounces@xxxxxxxxxxxxxxxx] On
Behalf Of John F. Sowa
Sent: Wednesday, August 31, 2011 11:41 AM
To: ontolog-forum@xxxxxxxxxxxxxxxx
Subject: Re: [ontolog-forum] Semantics of Natural
Languages (015)
Rich, (016)
Some comments: (017)
JFS
>> About a dozen years ago, I was talking with
Mike Genesereth,
>> who said "Lenat is probably the only one who
doesn't know that
>> Cyc has failed." (018)
RC
> That is the kind of thinking that all of us show
in one form or
> another. We seem stuck in our structured ways
after the first four
> decades or six, unable to overthrow the past
beliefs and institute
> new untried ones. (019)
Genesereth has been one of the strongest
proponents of classical
logic-based AI. He has been teaching at Stanford
for years in
close collaboration with the same people
(McCarthy, Feigenbaum,
Fikes, etc.) as Lenat. Any success stories from
Cyc would have
provided more attention (and funding) for all
kinds of projects
that used logic-based AI. But Mike G. was being
realistic. I
would qualify his comment, but I certainly
couldn't refute it. (020)
In my 1984 book, I tried to take a balanced view
of the strengths
and limitations of logic-based systems. My view
then (and with
more input since then) has been that logic-based
systems are
important, especially for applications to comp.
sci., but that
NLP systems must include logic-based approaches as
a proper subset: (021)
1. Large numbers of applications in computer
systems, database
systems, programming systems, and
hardware/software design,
require a foundation in formal logic. (022)
2. Natural languages can be used in very precise
ways (for
example, along the lines of controlled NLs),
but they
can also be used in very scruffy, very
informal ways. (023)
3. The overwhelming volume of NL speech and
documents use
highly informal, often ungrammatical, and
"innovative"
language. (I'm using "innovative" as a
neutral term
for what many people would call "incorrect".) (024)
3. I also agree with the comment by Alan Perlis
that you
can't translate informal language to formal
language by
any formal algorithm. (025)
4. I believe that you can interpret highly
informal language
by computer, but that you need to use huge
amounts of
background knowledge (i.e., extralinguistic
information)
to do so. (026)
5. Point #4 is acknowledged by classical
logic-based AI projects
such as Cyc. But they assume that you need a
long gestation
period that depends on hand-coded logical
representations
(e.g., formal ontologies and knowledge bases). (027)
6. The scruffies, such as Roger Schank, disputed
that claim
from the early days (1960s). But they didn't
have the
facilities for acquiring, storing, and using
such large
volumes of information. (028)
7. The hardware today is more than adequate to
store and
process the huge volumes of information
needed to support
point #6. One example is the IBM Watson
project, but
there are other projects that have achieved
comparable
success with more modest hardware resources.
The
VivoMind applications I summarized are among
them. (029)
> I don't see much of anything discussed about Cyc
past the
> precursors I mentioned anywhere in the public
literature;
> I'm not referring to tutorials about Cyc, but
about analyses. (030)
For the research publications, see (031)
http://cyc.com/cyc/technology/pubs (032)
For free downloads of the ontology and supporting
software: (033)
http://opencyc.org/ (034)
I believe that there are many useful applications
of Cyc and OpenCyc,
but I also believe that a different architecture
is necessary to
achieve something that could be called natural
language understanding. (035)
That is what I have been discussing in talks,
publications, and emails. (036)
John (037)
__________________________________________________
_______________
Message Archives:
http://ontolog.cim3.net/forum/ontolog-forum/
Config Subscr:
http://ontolog.cim3.net/mailman/listinfo/ontolog-f
orum/
Unsubscribe:
mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/
To join:
http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePa
ge#nid1J (038)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Config Subscr: 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 join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J (039)
|