cuo-wg
[Top] [All Lists]

[cuo-wg] Resolving Jim Hendler's Comments

To: common upper ontology working group <cuo-wg@xxxxxxxxxxxxxx>
From: "Schoening, James R C-E LCMC CIO/G6" <James.Schoening@xxxxxxxxxxx>
Date: Mon, 12 Feb 2007 13:03:15 -0500
Message-id: <5F6E70D8ED5D274F9D9A721485C0A46213EA593E@xxxxxxxxxxxxxxxxxxxxxxx>
CDSI WG,    (01)

1. Lots of good discussion, but let's refocus on resolving Jim Hendler's 
comments and updating the paper.      (02)

2. I would summarize Jim Hendler's comments as saying the paper doesn't give 
credit for the progress being made by the Semantic Web approach to semantic 
interoperability.  The attached paper is updated to give more credit.  I 
believe he admits this approach ("URI-based reference mechanism coupled with 
the standard for KR")is not yet ready for enterprise-wide implementation.  The 
paper is updated to list this approach as a 'Candidate Technical Solution.'  It 
rates the Semantic Web approach at Level-4.  Thoughts anyone?    (03)

3. I changed the readiness scale in the paper to the better-know Technology 
Readiness Scale. See
http://en.wikipedia.org/wiki/Technology_Readiness_Level.  Once again, this is 
far from a perfect model, but it is good enough to help explain how certain 
technologies are far from ready for implementation.     (04)

4. Below are excerpted and grouped comments, plus my shot at a response.  To 
respond, please cut only Jim Hendler's comments and add your response, 
preceeded by your last name.  Leave out Jim Hendler's comments you are not 
responding to.      (05)

5. At the bottom of this thread is the full text of the four messages sent by 
Jim Hendler, which may help put some of his comments in context.     (06)

6.  Jim Hendler comments and Jim Schoening responses:    (07)

a. Hendler Comment: 1) You make the mistake, which is common and being pushed 
by a number of strong players, of assuming that only with some kind of common 
semantic model can data be integrated... 2) it buys way too heavily into the 
assumption that no integration can take place without complete (and consistent) 
semantic agreement - 3).we couldn't ever succeed with any kind of "everyone 
must agree" (standard upper >ontology) ...      (08)

        Schoening Response: The paper makes no such claim, but does list 
'common upper ontology' as one candidate technical solution. The paper first 
makes the point that current technology can't achieve interoperability across 
large numbers of domains, then suggests further R&D is needed, then lists 
'candidate' technologies, but without claiming which is best, or even feasible. 
The paper has now beed changed to list Semantic Web Approach as a candidate 
technical solution.     (09)

b. Hendler Comment: ...buys way too heavily into the assumption that no 
integration can take place without complete (and consistent) semantic 
agreement.  ...that doesn't appear to be true in practice      (010)

        Schoening Response: The paper does not make such a claim.  The paper 
says semantic interoprability can be achieved in some cases, but not across 
large enterprises with many domain.  If there is practice to demonstrate 
enterprise-wide data interoperability, comments should cite such practice.          (011)


c. Hendler Comment: The ... argument you use as to why this is so (the scaling 
arguments) were very similar to the arguments made on why the WWW wouldn't work 
a decade or so ago.          (012)

        Schoening Response: Such arguments were obviously wrong regarding the 
web, but that is not a technical justification to dismiss them this time.    (013)

d. Hendler Comment:Think of it this way - if I could create a set of local 
mappings across a wide array of linked (but not all linked) data sources then I 
could indeed do cross domain integration.         (014)

        Schoening Response: Yes, one could do this for a narrow set of 
concepts, but not for all the concepts in all the domains within a large 
enterprises.     (015)

e. Hendler Comment:  when I ask Google to find things, I don't care if it finds 
all and only the right thing -I want it to take me closer to a correct starting 
point for exploration, and I don't demand (or in fact want) 100% precision or 
recall.     (016)

        Schoening Response: True, but Semantic Interoperability requires 
computers to understand search engine returns, which they can't, yet.    (017)

f. Hendler Comment: If you made this same argument by analogy for hypertext 
systems (lack of standards, disagreement in worlds views, different 
technologies) then you could clearly demonstrate that a world wide web of 
billions of documents linked to each other to be used in multiple contexts 
would be impossible (how would you find anything??) -           (018)

        Schoening Response: This analogy doesn't prove anything. The success of 
the Web is not a technical argument for the success of the Semantic Web. It is 
probably good justification for research funding, but is not a basis for 
claiming the technology can work, let alone is already working.      (019)

g. Hendler Comment:...that [argument that] has been used by many critics from 
the traditional ontology space (where that assumption is made) to argue you 
can't do data integration at a large scale with the approaches we're exploring.     (020)


        Schoening Response: By saying "...we're exploreing," this comment 
admits the Semantic Web approach is not mature yet, which is the main point of 
the paper.  The paper now lists this approach as a candidate technical solution.    (021)

h. Hendler Comment: I've said this many times in many contexts - Tim 
Berners-Lee (who never listened to those skeptics who explained why the Web 
wouldn't work) and Eric Miller and I addressed this issue in some degree in the 
paper at [1] ad of course I've written and talked a lot about this. So before 
you all throw out the baby with the bathtub, I thought I would mention that 
there are contradicting views            (022)

        Schoening Response: Not a technical comment.  No response needed.     (023)

i. Hendler Comment: you really missed a lot of the point of the Sem Web 
technologies and how they were designed precisely to provide the capabilities 
you say they don't.           (024)

        Schoening Response: I agree this is the intent of the Semantic Web, but 
this vision has yet to be achieved, so it is not ready for enterprise-wide 
implemention, which is the main point of the paper.    (025)

j. Hendler Comment: I certainly don't think they are the be-all and end-all, 
but they move us much further than you give them credit for -    (026)

        Schoening Response:  Agreed. Paper now gives more credit to progess of 
Semantic Web approach.     (027)

k. Hendler Comment: you completely miss the key aspect of these languages as 
opposed to previous AI languages, which is the URI basis - they are "webized" 
in a deep and important way - the linking of concepts (where ontologies can 
link to concepts in others) provides a mechanism you've largely missed.      (028)

        Schoening Response: Agreed. Paper has been updated to address this, but 
this does not impact the main point of the paper.    (029)

l. Hendler Comment:...they still haven't gotten it... The good news is the 
large players are beginning to get it (not only Oracle and IBM in DB systems, 
but MS and others are starting to use RDF DBs because they need the flexibility 
and the linking)           (030)

        Schoening Response:  This paper does not claim there is no progress 
toward the goal, just that technology has not yet achieved it.     (031)

m. Hendler Comment:  the key is, like the web, learning to live with 
inconsistency and ambiguity, rather than claiming you can't do integration in 
its presence.     (032)

        Schoening Response:     This paper does not claim we can't 'ever' 
achieve the goal, but rather we can't do it today with currently mature 
technology. Unless someone can show examples of Level 8 or 9 technological 
success, the primary assertion of the paper remains standing.          (033)

n. Hendler Comment: There's obviously no way to prove that ?X can be done with 
current technology, what I'm arguing against is your equally unprovable 
contention that it can't.            (034)

        Schoening Response:  The first statement admits technology has not yet 
demonstrated it can achieve wide-scale semantic interoperability. When large 
enterprises decided to emplement the Web, it had fully taken off and there was 
lots of evidence it was working and scaling up. We don't have anything close to 
that yet with the Semantic Web. Yes, there is movement in that direction, but 
not enough for large enterprises to successfully implement as a solution for 
data interoperability. The second statement misquotes the paper. The paper 
doesn't claim it can't, just that it isn't ready for broad implementation.     (035)

o. Hendler Comment: What I do believe is your arguments ignore significant 
aspects of current technology (esp. the Semantic Web work) that take it much 
further towards what you are claiming it cannot achieve than you seem to think.     (036)


        Schoening Response: Agreed.  Paper now gives credit. But this does not 
refute the paper's claim that the technology is not ready for broad 
implemenation, or in terms of the paper, the technology is not at Level 8 or 9.     (037)

p. Hendler Comment: I buy your argument if you are saying current technology 
defined as DL reasoners manipulating OWL assertions, but that's not what the 
Semantic Web is all about, the URI-based reference mechanism coupled with the 
standard for KR and other aspects is aimed exactly at scalability.          (038)

        Schoening Response:  The problem is actually worse than Jim Hendler's 
example (DL, OWL, etc). DoD believes it can achieve enterpirse-wide data 
interoperability with XML and metadata. Jim Hendler, in this comment, basically 
agrees enterprise-wide semantic interoperability cannot be achieved with DL 
reasoners manipulating OWL assertions. The second statement claims the 
Semantic Web approach is 'aimed' at scalability, which basically admits they 
haven't matured this approach yet. The latest version of the paper lists this 
approach as a candidate solution and advocates further work in maturing all 
candidates.      (039)


-------------------------------------------------------------------
Below is full text of Jim Hendler's 4 messages:  (Note:  A few non-relevant 
parts are snipped out.)    (040)

----------------------------------------------------------
    At 5:35 PM -0500 1/29/07, Jim Hendler wrote [to Jim Schoening]:    (041)

>James - Interesting paper, fwiw, I couldn't disagree more.   You 
>make the mistake, which is common and being pushed by a number of 
>strong players <<snip>>, of assuming that only with some kind of common 
>semantic model can data be integrated - problem is the same argument 
>you use as to why this is so (the scaling arguments) were very similar 
>to the arguments made on why the WWW wouldn't work a decade or so ago.
>Think of it this way - if I coudl create a set of local mappings across 
>a wide array of linked (but not all linked) data sources then I could 
>indeed do cross domain integration - not with full fidelity - but 
>that's where your mistake lies - when I ask Google to find things, I 
>don't care if it finds all and only the right thing - I want it to take 
>me closer to a correct starting point for exploration, and I don't 
>demand (or in fact want) 100%
>precision or recall.   I don't have time for details, 
>  <<snip>>
>  -JH    (042)

-------------------    (043)

At 10:33 AM -0500 1/30/07, Jim Hendler wrote [to SICoP]:    (044)


>I guess I should mention here something I told the author separately - 
>I don't agree with many of the conclusions of this paper, and think 
>there are some flaws - I don't have time for a detailed response - but 
>let me point out that if you made this same argument by analogy for 
>hypertext systems (lack of standards, disagreement in worlds views, 
>different technologies) then you could clearly demonstrate that a world 
>wide web of billions of documents linked to each other to be used in 
>multiple contexts would be impossible (how would you find anything??) - 
>the analogy isn't perfect, and there are some valid points made in the 
>paper - but it buys way too heavily into the assumption that no 
>integration can take place without complete (and consistent) semantic 
>agreement - and that's the part I cannot agree with, that doesn't 
>appear to be true in practice, and that has been being used by many 
>critics from the traditional ontology space (where that assumption is
>made) to argue that you can't do data integration at a large scale with 
>the approaches we're exploring.   I've said this many
>times in many contexts - Tim Berners-Lee (who never listened to those 
>skeptics who explained why the Web wouldn't work) and Eric Miller and I 
>addressed this issue in some degree in the paper at [1] ad of course 
>I've written and talked a lot about this. So before you all throw out 
>the baby with the bathtub, I thought I would mention that there are 
>contradicting views
>   -Jim H
>p.s. Actually, probably the best argument I made as to why we couldn't 
>ever succeed with any kind of "everyone must agree" (standard upper
>ontology) approach was in the original brief to the Director where I 
>convinced DARPA to invest in the DAML program  - so this isn't 
>something new, and those who've heard me at the Semantic Web in E-gov 
>conferences have heard this argument.
>
>[1] http://www.w3.org/2002/07/swint    (045)

------------------------    (046)



At 1:12 PM -0500 1/30/07, Jim Hendler wrote [to Jim Schoening]:    (047)

basically same thing basic argument as I put on the list, but a little stronger 
in my comments on upper ontology...  Sorry to weigh in so strong, but I think 
you really missed a lot of the point of the Sem Web technologies and how they 
were designed precisely to provide the capabilities you say they don't.  I 
certainly don't think they are the be-all and end-all, but they move us much 
further than you give them credit for - part of the problem is you completely 
miss the key aspect of these languages as opposed to previous AI languages, 
which is the URI basis - they are "webized" in a deep and important way - the 
linking of concepts (where ontologies can link to concepts in others) provides a
mechanism you've largely missed.   I realize I 
had the opportunity to weigh in to some of this stuff in SICOP, but I've had 
the debate with Leo and others ongoing for the past 6 years and they still 
haven't gotten it and I get tired of having the same arguments over and over.  
The good news is the large players are beginning to get it (not only Oracle and 
IBM in DB systems, but MS and others are starting to use RDF DBs because they 
need the flexibility and the linking) so I've given up on arguing in closed 
circles - I still write articles and give plenty of talks
  the key is, like the web, learning to live with inconsistency and ambiguity, 
rather than claiming you can't do integration in its presence.
<<snip>>    (048)

  -JH
-------------------------------    (049)

At 12:08 AM 1/31/07, Jim Hendler wrote [to SICoP]:    (050)

The goal of that paper is to claim (and later papers have more details) that 
the current technology is exactly aimed at achieving interoperability at the 
semantic level at a Web Scale, which certainly subsumes large enterprises 
with many domains.   I've published several 
papers on this ranging from vision papers like the Scientific American article 
and the agents on the semantic web (google scholar for Semantic Web finds 
these) and more technical ones - our technical papers are at 
http://www.mindswap.org/papers.    (051)

There's obviously no way to prove that ?X can be done with current technology, 
what I'm arguing against is your equally unprovable contention that it can't.  
What I do believe is your arguments ignore significant aspects of current 
technology (esp. the Semantic Web work) that take it much further towards what 
you are claiming it cannot achieve than you seem to think.  I don't have time 
to recap all the arguments here, but I think the bottom line is that I buy your 
argument if you were saying current technology defined as DL reasoners 
manipulating OWL assertions, but that's not what the Semantic Web is all about, 
and the URI-based reference mechanism coupled with the standard for KR and 
other aspects is aimed exactly at scalability.    (052)

I don't have time for a long email discussion on this right now, I simpl wanted 
to remind the forum that there is argument with your contention, and that not 
everyone agrees (both of which are self evident arguments). 
I leave it to the readers of your paper to think about these issues - that's 
all.
  -Jim H.    (053)













































Attachment: Data Interoperability Across the Enterprise 12 Feb-07.doc
Description: MS-Word document


_________________________________________________________________
Message Archives: http://colab.cim3.net/forum/cuo-wg/  
Subscribe/Unsubscribe/Config: http://colab.cim3.net/mailman/listinfo/cuo-wg/
To Post: mailto:cuo-wg@xxxxxxxxxxxxxx
Community Portal: http://colab.cim3.net/
Shared Files: http://colab.cim3.net/file/work/SICoP/cuo-wg/
Community Wiki: 
http://colab.cim3.net/cgi-bin/wiki.pl?SICoP/CommonUpperOntologyWG    (01)
<Prev in Thread] Current Thread [Next in Thread>