The mind reels.
The relationship between logical theories and the "real world" is based on the phenomenon of "ontological commitment" -- the decision to take certain metaphysical ideas to be true. It follows that a good formal ontology (in the knowledge engineering sense) is clear about what those commitments are. Pat's papers were very carefully based on a specific set of such commitments.
One of the problems with many "ontologgers" is that they have no idea when they are even making such commitments. They don't ask themselves whether there are other possible interpretations of their observations; they just write down what they think they see.
With respect to selling the idea of "AI ontology" to skeptics, I stated my position at one of the last DAML meetings: The development of "ontologies" is just an outgrowth of "information modeling" which is itself an outgrowth of "data modeling". This is evolutionary in the usual way of engineering disciplines -- first you figure out how to do something useful that works reliably over many variants, then you develop the theory for why that works.
[Begin unnecessary treatise]
Data modelers of the late 1970s and early 1980s were trying to make programs and databases that "worked the first time", by having thought through the problem of how to represent the world of interest in data forms BEFORE trying to write most of the code. So they abstracted the thinking from the code per se, but they ended capturing that model of the world ONLY in some implementation language form.
Information modelers of the 1980s and 1990s were trying to represent the world of interest in a way that was independent of the implementation language, and thus unburdened by its noisy pragmatic requirements. But they did not isolate their models from what OMG calls the "platform class" -- the kind of tooling they will use. That is why people who design relational databases and people who design XML data structures and people who design C++/Java programs came up with different models of the same problem space. The early and most successful group was designing databases, the later group was designing object-oriented programs, and the final (and least educated) group was designing XML schemas. And every <expletive> one of them insisted that their models, which were influenced, even transmogrified, by their implementation paradigm, were "natural". With the exception of ORM and Grady Booch's work, none of them had a formal theoretical foundation, as distinct from a
so-call
ed "formal semantics". The model ultimately meant: this (presumed) behavior of the world of interest will be captured in an implementation in this way. But these approaches were still a great leap forward -- they had the advantage of eliminating a lot of implementation noise in capturing and *presenting* an understanding of the domain.
In the same time frames, part of the AI world developed parallel modeling approaches, based on different intended implementation mechanisms, notably "frame logics" and "description logics". The difference was that these approaches had some formal theoretical foundation at their heart, because the implementation mechanisms did. "Frame logic" models are object models, to which a reasoning approach, rather than just a collection of subroutines, is attached. Description logic models are information models firmly founded in set theory and restricted by the requirements of a particular reasoning algorithm. These reasoning approaches are simply different implementation paradigms. BUT they are grounded in a formal theory, rather than just a reliable engineering practice, and that theory is the basis for the implementation paradigm. That is the difference.
OWL is not a great leap forward. It is the next step in learning how to model the world in way that can be used in a number of reliable engineering practices, without getting deeply into the nuances of those practices. OWL models are not "natural" either -- you can't say anything the extended tableaux reasoners cannot handle. But they are not formally interpreted by engineering practice -- the formal interpretation is a mathematical logic theory. To distinguish this kind of approach from the assorted forms of "information modeling" of the 1980s and 1990s, we use a new term "ontology". But it is just the third stage of evolution in modeling a world of interest for the purpose of automating some tasks in that world.
As Natasha Noy once observed, a great many OWL models are just UML models. Most of the classes are primitive, which greatly limits the value of the reasoning algorithms associated with description logic. All that means is that the would-be knowledge engineers are just learning how to do information modeling, and using OWL for the purpose (probably because it is a W3C standard, i.e. comes from the acknowledged source of all useful references for software engineering in the 21st century, as every illiterate software engineer knows). In the last 5 years, the UML folk have come up with formal logic models for UML interpretation as well (with the probable effect of invalidating the engineering interpretations of many existing UML models). This gives us an object-modeling language that has a formal foundation, much like the information modeling language ORM (in 1990). The difference here is that the formal foundation does not lead to the use of the model for automated reasonin
g. OWL
makes it possible to do that, even if many of its users are just learning to build information models, and thus do not make reasoning models in OWL. (Presumably their intended platform class is different.)
[end unnecessary treatise, which is probably inaccurate in several areas that John Sowa will point out]
The gist of this treatise is that "ontologies" are just somewhere on the upper end of an evolutionary curve in capturing the concepts in a world of interest in a machine-readable form, for the purpose of communicating among the stakeholders and automating some process. It is a part of the evolutionary process of developing a "computer science" -- a formal theoretical basis for software engineering. It is the current refinement of "world modeling" technology after 50 years of development, and practitioners use it with various levels of skill.
-Ed
-----Original Message-----
From: ontolog-forum-bounces@xxxxxxxxxxxxxxxx [mailto:ontolog-forum-
bounces@xxxxxxxxxxxxxxxx] On Behalf Of Christopher Spottiswoode
Sent: Tuesday, November 05, 2013 10:57 AM
To: '[ontolog-forum] '
Subject: Re: [ontolog-forum] Early use of the word 'ontology' in AI
John,
Thanks for the definitions of terms around the word "ontology".
But may I expand on my request in the final paragraph of my last email?
I had said:
" I am busy promoting ontology in a different forum at the moment. The
word seems to have some attraction there but I have discovered that, like
"semantic web", "ontology" is a sure turn-off out there to a significant
degree. So I am interested in both the history of the word and the current
experience of Ontolog folks in this labelling matter."
On the historical issue, though, Pat did say (in the quote in my email) that "I
wouldn't be [in favor of introducing a new term] if the term really were new
(and I resolutely ignored it when it was first introduced) but its no longer
new, [...]". That certainly seems to be an insistence that he was not the first
to introduce the term.
Then, the quote you attribute to Pat in your email below was in fact from me.
It was my own tentative exploration for a reading that rendered Pat's
statements consistent. I could surely learn from any attempt by this Forum
to spell out definitions of "logical theory" and "AI" that might clarify the
apparently diverse origins of the use of the term "ontology".
Thus the main definitions my exploration invited were those to help
relevantly distinguish the "logical theory" domain from that of AI.
I must confess I haven't studied Pat's 35-year-old papers on liquids and naïve
physics (though I do recall having ignorantly read about them at the time).
Nor would I characterize the present central themes of my own current
project as either 'logical theory' or AI. Ontology in that project is both most
strongly philosophical and of great technical value in IS/IT, and the word is
most useful in capturing that synergetic link. But I recognize that I do need to
gear up towards the relevant experts'
exploitation of the ever more emergent opportunities for both advanced
'logical theory' and AI in that project.
Ha-ha, yes, everyone can surely agree with your observation that "For
anybody who doesn't [know what they're talking about], all bets are off."
And one reason why I am still using the word "ontology", more or less in the
sense that ontologgers use it, despite my own inclinations (that I had
expressed in the sequel to the 2008 thread I had resurrected in my previous
email), is that whenever I can have a good conversation with sceptics, from
whatever field, within or outside IT, they usually end up understanding
surprisingly well at their respective levels and asking to be kept up-to-date.
But that experience doesn't resolve the marketing or propagandistic need
that lies behind my query, in the final paragraph of my previous post, as to
how fellow-ontologgers experience the problem of the ontology-sceptic.
Such a sceptic is usually ignorant, as you point out, and is adopting a very
superficial and usually self-serving position. So the obstacle to pre-empt
somehow is the first-glance though ignorant dismissal without the
opportunity for that "good conversation". For example, which of the one-
liners or elevator pitches of recent Summits seem to work best or most
often?
Any takers on such queries?
In the meanwhile I am busy taking my own approach to the whole issue, and
will in due course draw the Forum's attention to my project's present
deliverable. I am sure there are ontologgers who will be able to improve it
(even if by total replacement...). But I would much appreciate any
observations or suggestions in the very short term.
Christopher
-----Original Message-----
From: ontolog-forum-bounces@xxxxxxxxxxxxxxxx
[mailto:ontolog-forum-bounces@xxxxxxxxxxxxxxxx] On Behalf Of John F
Sowa
Sent: 05 November 2013 14:57
To: ontolog-forum@xxxxxxxxxxxxxxxx
Subject: Re: [ontolog-forum] Early use of the word 'ontology' in AI
Christopher,
In that note, Pat and Bill were discussing some issues about proper usage,
but not the question of first occurrence of the word 'ontology' in the AI
literature. Note Pat's brief summary:
PH
But the above concerns the term's use as a synonym for "logical
theory", whereas the present thread is addressing its use in AI.
In any case, there is as much confusion about the word 'theory' as there is
about 'ontology'. And the term 'logical theory' is less common (199,000 hits
on Google) than 'formal theory' (300,000 hits).
I'd say that every ontology is a theory. But to be an ontology, a theory must
have an additional claim about the existence of the entities that its variables
refer to. I'd also add the distinction of formal vs informal theories and
ontologies:
Theory: A systematic set of assumptions and their implications.
Ontology: A theory about what exists in some domain.
Comment: Note that a theory or an ontology can be stated in ordinary
language. But it must be systematic. An observation that there is a robin in
your backyard is not an ontology. But a detailed theory about birds and their
characteristic properties and relations could be called an ontology.
Formal theory: The deductive closure of a set of axioms stated in some
version of logic.
Formal ontology: A formal theory about what exists in some domain.
Comment: The adjective 'formal' as used in both philosophy and computer
science implies the use of a precisely defined notation for making statements
and drawing inferences -- i.e., a logic.
I believe that these four definitions correspond quite closely to the usage in
both philosophy and computer science -- at least by people who know what
they're talking about. For anybody who doesn't, all bets are off.
John
__________________________________________________________
_______
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
__________________________________________________________
_______
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