> On Tue, 2012-02-21 at 10:28 -0500, John F. Sowa wrote:
Leo, Hans, and Rich,
We all agree that common semantics is a prerequisite for any kind of
communication among people and/or computers. I would also add that
we need some common semantics even when we communicate with our
beastly friends and foes.
Questions for this forum: How much common semantics is required?
How should it be encoded? Is it primarily task-oriented semantics
that changes from one transaction to another? Should it be encoded in
a large shared general ontology? In some collection of task-oriented
modules or microtheories? In some implicit or procedural knowledge
encoded in procedures (for computers) or in habits (for people and
other animals)? In some mixture of formal logic, informal natural
languages, procedures, habits, or miscellaneous?
And the thorniest question of all: How do we accommodate the trillions
of dollars of legacy software in daily use for mission-critical systems
that won't be replaced for decades to come?
> I have only a dim understanding of finance, but I do recall a finance
> professor emphasizing the point that sunk costs are immaterial to future
> spending decisions. So past investments in software are of no concern. A
> rational CFO will only ask how much return he will get from the next
> dollar he spends. The question comes down to putting your enterprise
> intellectual property into proprietary OIDs (application-specific object
> identifiers) and all the machinery it takes to make those OIDs useful,
> or putting it into URIs (and RDF triples)
You have two different things here. Moving to triples is a step far
backward for people who have their information in databases which have
rows and columns. It is an awkward, but doable method of sending data
which naturally falls in non-sparse matrices.
If the Semantic Web had come up with a better syntax than triples, it
would be far more used, imho.
> and get much of the caring and
> feeding of them for free so you can invest more of your intellectual
> resources into getting the most out of your cool URIs. I think it's
> clear that a dollar spent on URIs will benefit you more than a dollar
> spent on OIDs (for any but the most ephemeral data).
But how many times as many dollars does it cost to create a system using
URIs? Certainly enterprises would not wish to spend 20-100 times as much
for data storage by using URIs. The increased bandwidth would also be a
great cost.
> And because of the
> multiplier power of linked data, you will obtain increasing marginal
> returns from future dollars spent on URIs.
The multiplier power of linked data has to be weighed against the
lack of context of data coming from elsewhere. Is a company supposed
to trust its profits on data that may come from its competitors or
from someone trying to sabotage their system? ... or even data that
is obsolete, or was entered by someone with a different concept of
what a linked data term meant?
> The enterprise data that is currently locked up in OIDs must of course
> be put into URIs and triples,
"Must"? Who decrees this? Why must they force their round pegs, eggs,
and spider webs into square holes?
> but that is bread-and-butter work (after
> the highly demanding work of creating the target model). Then there is
> certainly some valuable functionality the software vendor has provided
> in exchange for locking up your data. That is an opportunity for
> software-as-a-service providers to step up and build equivalent
> functionality for open data sets.
How many corporations want to put their corporate data in open data sets?
> <snip>
Recommendation: Instead of developing "proactive" standards for
ontology, I suggest that we note the law of standards: examine
what actually works, harmonize the best practices, and build the
new additions as extensions to the de facto standards.
> I'm reminded of the Dickens Pickwickian who thought he would write a
> paper on "Chinese Philosophy" by reading in the encylopedia under
> "China", and then under "Philosophy", and combine the two. I don't know
> how much harmony we would find by reading under "Description Logics",
> "ISO 10303", "ISO 15926", "SysML", and other nominees for best
> practices.
> I agree it is normally not prudent to get ahead of established practice.
> But I don't consider much of past practice (in the IT realm) to be worth
> saving.
Those who run their businesses on it might wish to maintain their existing
software until something that can be proved to be better to such an extent
that it is worth the cost of switching over comes along. The Semantic
Web obviously isn't it -- since corporations have not flocked there.
> Keep parametric solid modeling and discard the rest of IT's
> contribution to product and systems engineering, and we'll be better off
> because then we could see clearly.
I question this. I would certainly suggest learning from the current
contributions.
> We could take a fresh look at what
> the designers, builders, and users of systems actually *think* and *do*
> (and *why* they do those things), and maybe we could do some good for
> them.
Here, i agree.
-- doug f
John
_________________________________________________________________
<SNIP>
Regards,
--Paul