Re: Arcs

On 22/10/10 01:24, Andrew S. Townley wrote:
> On 21 Oct 2010, at 12:24 PM, Murray Altheim wrote:
[...]
>> Hi Andrew,
>>
>> Not really. As a species we've been trying to communicate with each
>> other since we could grunt, and none of the issues inherent in that
>> communication have gone away. What has changed is the availability
>> of tools that permit at the very least the complete availability of
>> documentation of what it is we're attempting to communicate, i.e.,
>> the ability to fully spell out (literally) the definitions of our
>> terms and relations in ways that permit discovery and analysis, to
>> both human and machine. These systems are still by and large horribly
>> primitive, but they exist for the first time. We can largely credit
>> Roy Fielding for this.
>
> I'd agree with the hat tip to Roy for sure.  I guess I wasn't very
> clear in my own semantics, but I was actually referring to all of
> the AI work and natural language processing (*that* NLP) inference
> work that's been going on during that time.
>
> It's an immensely complex problem, so solving it will take some
> time for sure.  I just see that there's two main camps for trying
> to address the "getting software to recognize owner flags" part:
> the machine-centric and the human-centric approach.

Yes. Unfortunately, those two camps aren't too clearly delineated
either by each other or by themselves. E.g., the whole use of
description logics (or any formal logic) by people rather than
machines, or by machines designed by people who aren't aware of
the ontological commitments of DL, strikes me as problematic at
best, dangerous at worst. Though I suppose the use of the systems
by those who paid for them is (hopefully) more informed.

> The machine-centric approach is what I mentioned above, based on
> formal semantics and including people like Sowa and the whole
> ontolog community (broad generalization alert!) as well as many
> of the DARPA dollars you mentioned.  This approach very much is
> focused on trying to encode semantic discovery into systems so
> that they can interpret these owner flags however said owners
> would like to state them.  All the better if they're in a
> controlled vocabulary like RDF+OWL of some variety, but hey,
> that's OK, right? ;)

My experience with that community is that a very small percentage
have any idea whatsoever of the epistemological minefield they're
walking in. A few I've run into don't even want to know as it
breaks their toys.

> (BTW, I don't have anything against John Sowa, and I have a lot
> of respect for his work.)

I don't think I'd be doing John any disservice by saying that his
greatest contribution to KR is probably as its historian,
secondly as its greatest teacher. But he is one of the more
informed and therefore cautious. He's a Peircean I believe because
he understands the issue of context as central (where in C.S.
Peirce it has a rather different name and more profound implication,
not to digress).

> The human-centric approach to me is much more exemplified by
> Topic Maps and was one of the main personal draws when I
> stumbled into this space in 05/06 even though I'd been walking
> in that direction without realizing it for several years before.
> However, Topic Maps also seems to take much more of a "train
> the humans" to express things in ways that are more easily
> processed by a system.  PSIs are a very obvious example of this
> in action.  Whether it strikes a good enough middle-ground is
> obviously a bigger discussion, but I think it does several
> things better than some of the alternatives.

One of the things I like about Topic Maps is that the ontological
commitment is almost zero. It's just a graph. That's why I have
so much difficulty with more recent interjections of semantics
that break that zero, like reification. To me that blows the
model by trying to capture something waaaay above the level at
which ISO 13250 and XTM 1.0 operate. But some very intelligent
people have put a lot more thought into that than me; perhaps I
simply don't understand the whys and wherefores well enough, as
I've not had the luxury of the face to face contact I did back
then. Travel budgets and all.

>>> Funny how semantic software that was supposed to eliminate
>>> interoperability problems has caused a new batch of interoperability
>>> problems. Some of it is a direct example of Murry's "flag planting"
>>> but some is that there's always a domain boundary people hit or
>>> adaptations that need to be made to make applications efficient
>>> in a particular context.
>>
>> I think that most of these problems are simply poor design. There's
>> no way to legislate good design, nor keep software designers from
>> jumping into solutions before they understand the actual problems,
>> nor from jumping ignorantly or fully cognizant of the rules they're
>> ignoring.
>
> I'm not sure I agree with you about "most", but certainly a good
> few.  I've a few pet peeves about "good design" that I won't
> bring into this discussion, but I think that as it becomes
> easier to create  functionally useful software where design is
> in many cases optional as long as it works, this problem isn't
> going to go away, and we're standing on the beach saying
> "go away" to the oncoming Tsunami.

I think we generally agree on this -- there are *so* many systems
that at first glance work ("functionally useful") but are
horribly broken underneath the hood.

> Real innovation always happens on context boundaries, and, let's
> face it, you're going to get it wrong more often than you get
> it right.  The other thing is that there's 10M opinions, most
> of which are commentary on probably 10 core principles.  The
> trick is to express those principles as clearly, concisely and
> accessibly as possible so those wanting to avoid some dead ends
> have obvious places to look.  *then* they can choose to ignore
> the rules if they (think they) fully understand them and the
> implications of so doing.
[...]

Certainly, we must allow enormous license in order to explore
and make mistakes. But looking back over the last century we
should also consider the big mistakes that have been made, but
few ever look back at all. That you're aware of Sowa (and I hope
you've had a chance to read his KR book, it's excellent as an
introduction) suggests you are interested in grounding your
work. That's a good thing. I am as well, and whether either of
us are successful is another matter, and we need to be willing
to make mistakes. I'm only advocating for those mistakes to be
informed by past ones.

>> To my knowledge, nobody has gone down this path, but technology can
>> also provide for *dialogue* in dynamically negotiating agreement.
>> Normally it's just the fudge factor (I have never seen a duck nor do
>> I know what a duck is, but I've read the documentation and I think
>> it sounds like a duck, walks like a duck; must be a duck).
>
> Actually, I wasn't speaking of domain knowledge silos in this
> case, rather of representation technologies and the bodies of
> knowledge (and underlying assumptions) that go with them.  Is
> what you say is the "same" really the same as what I mean when
> I say "same" because of the perspectives used to determine sameness.

Yes, the ontological commitments of those systems/technologies.

> This is also the crux to me of the difference between the
> machine-centric and the human-centric "schools of thought" I
> mentioned above, and it depends ultimately on how much utility
> a human can get by applying the "fudge factor" you mention.
> In fact, aspects of my implementation try and codify specifics
> about the "fudge factor" applied in making these determinations.
> Potentially, it could be useful for the types of negotations
> you describe.

Absolutely. One of the more productive areas of computational
linguistics research has been in statistical analysis of large
bodies of texts rather than small scale modeling. Google comes
to mind...

>>> I totally agree that more emphasis in this area would be very useful!
>>
>> One of the more interesting areas of research in AI/KR is context,
>> with at one point yearly conferences solely focused on the subject
>> (e.g., [1]). Topic Maps have a rather primitive concept of context
>> in the use of 'scope' that could be expanded, possibly within the
>> existing model, or perhaps we need to elaborate.  We should be
>> paying more attention to the results of that research. I seem to
>> remember David Kirch and Patrick Brezillion as central figures,
>> people to check out.
>
> I wasn't aware of those names, but thanks for mentioning them.
> I also wasn't aware of the conference, but the whole area of
> context and how to represent it sufficiently to be useful - in
>  another context, of course - has gotten a lot of thought cycles
> from me over the last several years.  I certainly agree that
> the scope stuff is, while useful in limited cases, one of the
> weaker areas I see in the whole Topic Maps space.

I don't think of it as weak but as the graph component necessary
to model a much more complicated representation of context. I
don't think we'd really need much more than TM scope to do that.
Scope is after all just another property.

> Information is data in context, and without adequate ways to
> represent and convey that context, you're open to lots and
> lots of issues.  Obviously, all of us deal with this problem
> every day in many different aspects of our lives.

And humans are remarkably able to resolve ambiguity, where
we aren't remotely close to designing machines to do that.

> You said "at one point" above.  Are you aware of particular
> ongoing efforts beyond the names you mention?

I am, but I'm *supposed* to be packing for a backpacking trip
this weekend and this thread has been happily keeping me from
doing that more mundane task. But if you ping me on or off the
list when I get back (next Tuesday), I'd be happy to dig some
references up.

>> I also think it behooves everyone to do a bit of epistemological
>> homework before jumping head-deep into the pool of modeling. It's
>> a deep pool and the bottom is littered with the bodies of those
>> who have gone before us and had no ability to swim. I won't name
>> names, but as a hint they had lots of DARPA money.
>
> I'd strongly agree with this as well, however, you're going
> to have to deal with the collision of "those who know how we
> know" with "those who know X" and there's a lot more of the
> latter than there are of the former.  Gets back to the points
> earlier that you and Patrick raised, but the perfect model is
> useless if it isn't usable in the wild by real people on a
> daily basis.  I know you know this, but that's one of the
> biggest challenges I see in this whole space.  We need stuff
> for the masses that helps meet the epistemological criteria,
> and I think we're still a long way from many examples of that
> kind of system.

I really couldn't agree more, and pedagogy, best practices
documentation, etc. is a good place to start. It is more a
human than a machine issue right now. We need to educate our
software developers and business analysts.

Perhaps this is a good place for me to drop out for awhile,
as I really must finish my packing. But I have enjoyed this
conversation (thank you) and I'd like to pick it up again
when I return if things haven't moved on too far by then.

Cheers,

Murray

...........................................................................
Murray Altheim <murray10 at altheim dot com>                       = =  ===
http://www.altheim.com/murray/                                     ===  ===
SGML Grease Monkey, Banjo Player, Wantanabe Zen Monk               = =  ===

       Boundless wind and moon - the eye within eyes,
       Inexhaustible heaven and earth - the light beyond light,
       The willow dark, the flower bright - ten thousand houses,
       Knock at any door - there's one who will respond.
                                       -- The Blue Cliff Record