On Oct 6, 2009, at 11:58 AM, Joe Collins wrote: (01)
> Dear Ed,
>
> Could you please specify GUM (a NIST pub, "Guide to Units and
> Measurement, or
> something?).
>
> To quote your signature message
> "We must strive to make things as simple as possible, but no simpler."
> - Albert Einstein
>
> In this case, regarding uncertainty, the question is, what
> complexity is
> necessary? As you point out, a theory of uncertainty supporting some
> reasoning
> capability could take significant work, but simply capturing
> expressions of
> uncertainty is easier. (02)
Not much, given that the a constraint upon any formalization we write
must be that it *could* be used by reasoning engines. (03)
> The two most significant expressions that come to mind
> are the interval and standard deviation. These meanings are pretty
> common; even
> universal. (04)
Intervals are easy, yes. I have never seen a logical formalization of
the meaning of standard deviation. I am sure it could be done, but it
would be complicated. (05)
>
>> But the grocer's scale reads 1.0, 1.1, etc. and the needle is
>> between 1.0 and
>> 1.1. What does that mean? Is it OK if he presumes that it is
>> accurate
>> enough to satisfy both of you that he is honestly selling you 1
>> kg? Once you
>> start adding 'uncertainty' ideas, we bring in the question of when
>> the scale
>> was last calibrated. Neither of you wants to think about that.
>
> In this case, the grocer will often say "It's between 1.0 and 1.1,
> is that OK?".
> At my grocer's, the weights and measures calibration date is affixed
> to the
> scale (by law in many locales). While I may not examine that sticker
> when buying
> lettuce, you I certainly may when buying truffles.
>
>> (4) We don't want to confuse tolerance with uncertainty in the
>> model. A
>> specified quantity value has tolerance (and usually ignores
>> uncertainty,
>> sometimes with serious consequences); a measured quantity value has
>> only
>> uncertainty. An abstract ideal 'quantity magnitude' has neither.
>> It is the
>> use or source of the quantity value that acquires the tolerance or
>> uncertainty property.
>
> I agree we shouldn't confuse tolerance with uncertainty.
>
> Like functions that are non-computable, the "true" value of a
> physical quantity
> is non-measurable and therefore not knowable. (06)
Even if true, this is irrelevant to ontology engineering. Ontology is
not epistemics. The ontology does not set out to describe what can be
known. It sets out to describe the reality described by the knowledge.
Unless you want to argue that because we cannot know a true value,
that therefore no true value exists - a coherent point of view, but
not one conducive to communication or normal conceptual description -
then please don't put scare quotes around 'true'. Electrons do have a
true rest mass, even if we can only measure it to a certain precision. (07)
> We only know the values of
> quantities by measurement and we always measure to some limited
> precision.
> Tolerances, also, are always non-zero. The source (via measurement)
> or use
> (e.g., tolerance) are the sources of any knowledge we have on physical
> quantities: we don't really know any more about "true" or "ideal"
> values. (08)
We still however assume that they exist, and our science and
engineering and trade rules and laws and our ordinary communication
all are based on this presumption. So our ontologies should speak of
these things that we all assume to exist. (09)
> I
> don't see how to relegate the representation of uncertainty to a
> microtheory: it
> is an essential part of the thing we claim to be knowledge. I'm not
> convinced
> that the interpretation is necessarily specialized, either.
>
> If as you assert:
>> If we go only one
>> step down that path, we will have a model that has a nearly
>> undefined slot
>> for "uncertainty", which means that any two uses of it are likely
>> to be
>> semantically incomparable. If we go further, it will take a few
>> years to get
>> a correct ontology in the area.
>
> I think the more important questions are "Must it be done?" and "Can
> it be
> done?" rather than "How long will it take?".
>
> My answers are "Yes", "Probably", and "I don't know", respectively (010)
and mine are: no, not in the near future, and God alone knows, but
people have been trying for several decades and havn't got very far. (011)
>
>
>> Yes, in the abstract, 'uncertainty' is well-defined. It is the
>> expression of
>> uncertainty that is interesting. Presumably we would include the
>> concept so
>> that uncertainties can be expressed. Would you be happy with a
>> 'quantity-value has-uncertainty simple-quantity-value' property
>> (with the
>> basic SI definition = the maximum amount by which 'quantity-value'
>> may differ
>> from the actual value)? What about conditions, e.g. "at standard
>> temperature
>> and pressure"?
>
> So, when is something completely unambiguous? Almost never. (012)
? Why are you talking about ambiguity? That is yet another distinct
topic (apart from probability and tolerance). (013)
> How much qualification is enough? As much as it takes for the user
> to be happy.
> Does the user always specify what it takes to make him happy? No.
> I think that an expression of uncertainty is a first level of
> qualification.
>
> So, my wife hands me a grocery list. It says "milk, bread, eggs". (014)
OK, already this example is irrelevant. Your problem here is
interpreting natural-language in a social context. This is about as
far as it is possible to get from anything to to with ontology
engineering. (015)
Pat (016)
> Suffice it to
> say that my wife is not easily trained to my expectations: she never
> specifies
> amounts, let alone a range of acceptable quantities. I do like to
> keep her
> happy, though, and, as important, I don't want to have to repeat the
> shopping
> trip daily.
> I have learned that, having 3 kids, a quart of milk is insufficient,
> and ten
> gallons is excessive. I've even narrowed it down better than that.
> Perhaps she
> has trained me to her expectations.
>
> I am happy to report, however, that, even after untold man-years of
> effort,
> computers are dumber than I am (stay tuned?).
> I'm pretty sure that if computers are ever able to deliver on such a
> request,
> then to keep my wife happy they will have to have a rudimentary
> capability to
> process uncertainty. Until they can, I still am useful. Perhaps I am
> trying to
> put myself out of a job?
>
>
> R/jbc
>
>
> --
> _______________________________
> Joseph B. Collins, Ph.D.
> Code 5583, Adv. Info. Tech.
> Naval Research Laboratory
> Washington, DC 20375
> (202) 404-7041
> (202) 767-1122 (fax)
> B34, R221C
> _______________________________
>
> _________________________________________________________________
> Message Archives: http://ontolog.cim3.net/forum/uom-ontology-std/
> Subscribe: mailto:uom-ontology-std-join@xxxxxxxxxxxxxxxx
> Config/Unsubscribe: http://ontolog.cim3.net/mailman/listinfo/uom-ontology-std/
> Shared Files: http://ontolog.cim3.net/file/work/UoM/
> Wiki: http://ontolog.cim3.net/cgi-bin/wiki.pl?UoM_Ontology_Standard
>
> (017)
------------------------------------------------------------
IHMC (850)434 8903 or (650)494 3973
40 South Alcaniz St. (850)202 4416 office
Pensacola (850)202 4440 fax
FL 32502 (850)291 0667 mobile
phayesAT-SIGNihmc.us http://www.ihmc.us/users/phayes (018)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/uom-ontology-std/
Subscribe: mailto:uom-ontology-std-join@xxxxxxxxxxxxxxxx
Config/Unsubscribe: http://ontolog.cim3.net/mailman/listinfo/uom-ontology-std/
Shared Files: http://ontolog.cim3.net/file/work/UoM/
Wiki: http://ontolog.cim3.net/cgi-bin/wiki.pl?UoM_Ontology_Standard (019)
|