Re: EDI with real semantics
nitin@sybase.com (Nitin Borwankar)
From: nitin@sybase.com (Nitin Borwankar)
Message-id: <9408122152.AA10219@trishul.sybgate.sybase.com>
Subject: Re: EDI with real semantics
To: edi-new-request@tegsun.harvard.edu (Scott M. Dickson)
Date: Fri, 12 Aug 94 14:52:28 PDT
Cc: cg@cs.umn.edu, srkb@cs.umbc.edu, edi-new@tegsun.harvard.edu
In-reply-to: <9408121924.AA23508@ontek.com>; from "Scott M. Dickson" at Aug 12, 94 12:32 pm
X-Mailer: ELM [version 2.3 PL0]
Content-Length: 7689
Sender: owner-srkb@cs.umbc.edu
Precedence: bulk
In your message you, Scott M. Dickson, graciously said
[...]
> However, something needs to be done in the meantime;
And this is the crux of the matter !!
We need some solution for problems "in our face" today.
For those using X.12/EDIFACT the "problem" is how to use
Internet infrastructure to extend the reach of their implementations.
Dave Crocker is very competently leading the ietf-edi effort on
this.
But for those of us not interested in an X.12 approach and in the
need of forms-compatible solutions, the "problem" is how to do
structured data exchange - today. Not a couple of years later.
>standardizing the
> legal interpretations of transactions, extending them to forms (i.e., "This
> form contains data for this type of transaction [which has a defined legal
> interpretation]"), tagging form data elements, and a protocol for automated
> negotiation of transaction data can all be done without waiting for the
> results of knowledge representation and exchange research.
Hooray!!
My sentiments exactly.
[...]
>
> They're formally defined, based on the ontology and whatever formal
> definition primitives you have (usually some variant of first-order logic).
> Getting a good ontology so that the wide variety of things referenced in
> trade transactions can be constructively defined from a manageable set of
> ontological primitives is "tricky".
Can the ontological primitives be stated in a crisp yes-no fashion ?
Are there items which have clearly defined memberships in crisply
defined sets ?
A counterexample is an inventory model for a store that sells clothes
for "big and tall" people.
Does the set of all "big and tall people" have well defined boundaries.
Is one short below 6 feet and at 6 feet 1" become "tall" ?
The inventory, ordering and re-stocking patterns for this
will be strongly affected by this model and the customer-set.
First-order logic will not be enough for sets with non-crisply defined
memberships and questions that have fractional degree answers.
See my detailed comments further down below on first order logic, intelligent
agents and business models.
[...]
> >Another is the related enterprise-integration system
> >of Ontek, Inc. in Laguna Hills, Cal.
>
> Thanks for the plug.
> Ontek's is a somewhat different from the others, and the CSMF effort has a
> little SUMM (Semantic Unification Meta Model) thrown in, along with
> modeling languages like IDEF, NIAM, etc.
Scott, could you give us a brief summary of what Ontek is doing ?
> One can purchase a great variety of things, only some of which are physical
> artifacts. PDES/STEP has focused on the geometric and material description
> of physical artifacts. One can also purchase services, corn in the future,
> real estate, hours of a person's time, insurance, a business, a patent, the
> fulfillment of a requirement, ad infinitum. The description of product in
> the sense of thing to be purchased will require a very rich ontology.
One can also purchase information, which has a whole universe of sub-categories
in itself.
I, personally, am interested in forms-edi because it is a natural query
interface for structured information bases that need to be queried in
ad hoc ways. The information result set could be considered a "product".
How do all these AI/Knowledge based schemes deal with meta-data about
*information as a product*, if at all ? I suspect - not at all.
Information-as-a-prodcut is increasingly becoming a larger segment
of the ( US ) economy.
How will these structured information-as-a-prodcut transactions
be modelled in the knowlede based schemes ? Or is the knowledge only
about transactions involving concrete "things".
> This means that the ontology will have to provide for uncertainty,
> uncertainty will be part of the definitions of the constructs and any
> inference mechanisms will have to be tolerant of uncertainty. Law is a
> combination of written code, tradition and often conflicting
> interpretations of those.
And how does one deal with uncertainty in a sensible way with first-order
logic alone ? Can one do it at all ?
This is why I mentioned, in an earlier message, the work of Zadeh and Kosko
which seems to deal with these issues very well. First-order logic falls flat on
its face when set membership is not crisp ie not yes-no.
A recent practical application to real business modelling problems is
detailed in "A Fuzzy Systems Handbook" by Cox. It has source code in C.
> >I would be very interested to see what the "new-EDI" thinkers (and
> >the EDI Establishment) have to say about this kind of advanced
> >programme for EDI.
That it may not be enough.
If it is based on first-order logic it will suffer from all the
problems associated with combinatorial explosion for rule based systems
in real world AI applications. Performance of such systems - time to
process rules - will always be an issue.
Conflicting rules ie ambiguity, will be handled in a highly implementation
dependedent way since "A *and* not-A" is not allowed in first-order logic.
Implementers will make possibly ad-hoc judgements on how to resolve
conflict thus creating inconsistent world-views.
Ambiguity is tractable in the "fuzzy logic" schema where "degree of membership"
in a set can have fractional values and degree of ambiguity,
= measure ( A intersection ~A ), can be non-zero.
Fuzzy logic deals with "linguistic variables" like "tall" whose
values are not single numbers or crisp sets, but are "fuzzy sets".
Thus a 6 ft 6" person has a "tallness" value ( say ) quite-tall, and a degree
of membership nearer to 1 in the "tallness" set.
This is context dependent - when applied to a set of basketball players
the same person may have a value of "average" for the linguistic
variable "tallness", which in turn translates to a number.
While decisions are being made fuzzy sets are used, and when an output
value is required, the centroid of the set is computed to map to a single
number.
The use of linguistic variables allows a smooth translation from information
derived through interviewing experts to system models using
linguistic variables and fuzzy logic.
This is not an idle academic issue. The Japanese are using this
technology to create "smart appliances" that will adapt to surrounding
conditions. A vacuum cleaner will adjust its speed according to the degree
of roughness of the carpet and the length of fibres etc.
Really useful "intelligent agents" should be able to do the same
in the presence of ambiguity.
I can't see how agents can be "intelligent" if they use only first order logic.
If you really want to model business environments you need to deal
head-on with ambiguity and use it constructively to make judgements.
Not hand-wave it away by the use of a model that denies it's existence.
The approach of Zadeh/Kosko and applications by Fox are real world
examples of operational business modelling using these methods today.
>
> It would be helpful to put together a crude time line, outlining possible
> near-term and far-term improvements.
Yes, it's definitely clear that all the things discussed on this list
are relevant *in their own time-line*.
Could you begin the process by suggesting an outline that could be
used as a starting point. Since I have been ( and possibly will be )
a vocal advocate of one of the points of approach ( forms-edi),
I would rather have an "outside observer" initiate this so that my
biases are not imposed on the process at this critical point.
I am assuming other vocal advocates
will also take a similar even-handed approach at this stage.
Hint. Hint :-)
So ? Want to give it a shot ?
Nitin Borwankar,
nitin@sybase.com