Edited
September 17, 2004
Anticipatory Web
5/4/2004
4:30 PM
From anonymous, to Paul Prueitt
(edited to shorten. Comments by Prueitt in italics indented)
Paul,
Most ontologists still laugh at the thought of a common upper ontology, of the type proposed by Mike and others at SICoP.
We agree. Some serious work has
occurred at Cyc Corp on upper ontologies, in fact it
is this work that may be responsible for the term “upper ontology”. One problem is that this work has not shown that
a combination of upper ontologies and what they call “micro-theories” solves
what I will call the “problem of closure”.
The problem of closure occurs when a formal representation of human
knowledge is fixed and the situation changes or the interpretation of terms
changes so that the now fixed and operational formal representation is
mismatched to the real time situation. One ends up trying the fit a round peg
into a square hole. One way to characterize
the limitations of Semantic Web type ontology is to discuss the formal
limitations of Hilbert mathematics and classical logic.
There is market interest in automated taxonomy and automated ontology
construction. The interest is
instrumented via a correspondence to both back of the book indexes and formal
taxonomy. InOrb Technologies has
developed something similar, to “upper ontologies”, called upper taxonomy. It manly involves the selection of 200 – 400
words whose contextual meanings “cover” the
anticipated subject matter in a domain of discourse. The selection process can be quite arbitrary or not, and result
in poor selection or excellent selection.
One way to reduce the impact that the closure problem has with an upper
taxonomy is to allow new terms to be introduced by community members and to
allow the not-used terms to be dropped.
The active role of communities in the maintenance of an upper taxonomy
needs to have reconciliation technology and processes, such as those pioneered
by Schemalogic Inc.
A categorical difference between ontology and taxonomy may be the
origin of the controversy over what ontology means in the natural
sciences. Ontology expresses as
substance and form, and in this way has a similarity to how terms “gene” and
“phenotype” are used. A gene expresses
as phenotype. In many of the
theoretical discussion on gene expression, the role of the environment in the
expression makes it reasonable to suggest that the gene does not exist except
through the interaction between substances and the environment. The gene has a replicator function and has
some level of substance, but the exact nature of the gene is technically
complex. Ontology should have this same
technical complexity, and in my work it does. But the type of ontology that is being used
by most, or all, members of SICoP does not.
More can be said on this, but it takes a while to understand how
important the notion of stratification
is.
Stratification is one of those notions that are simply rejected, by
most members of SICoP, as being non-sense or not understandable, as in the previous posting.
However, from the notion of stratification one is able to develop a
mechanism where underconstrainted assembly of atomic elements is given
additional constraints in real time.
Human-centric Informational Production can then be developed for the
following reasons:
1)
Strong forms of artificial intelligence, and strong forms of cognitive
engineering are set aside so that real time manipulation of information and
meaning can occur.
2)
Human cognitive acuity is allowed to create action perception cycles in
real time by an individual whose is both engaged and interested in a specific
issue in real time.
In the physical theory supporting the Conjecture on Stratification,
science does observe specific interactions between (atomic) substances and the
environment. This is clearly what is
occurring in physical chemistry.
Stratification may be how memory recourses are separated from
anticipatory (orienting response mechanism).
This separation is clearly seen in the functional differentiation of
anatomically different regions of the brain.
For example such functional separation is discussed by Karl Pribram in “Languages
of the Brain” (1971) and in “Brain and Perception” (1991). The issue of separation of “memory” and
“anticipation” is not easy to be precise about; because awareness of memory is
induced by anticipation and the consequences of this mutual induction is the
formation of the aware state (or a “memory”).
(See last section of the PowerPoint developed to discuss the Readware Provenance ™
anticipatory HIP software.)
Again, the literature on this is referenced. The key scholarly field is called “ecological physics” which is a
sub discipline of psychology following the works of J. J. Gibson on action –
perception cycles.
The real time emergence of a situationally specific ontology, or
taxonomy, is what is desired in the ideal case. The sharing of ontology within communities of practice can be
done in this context.
In this context one would never have an “upper” ontology in the sense
used by Cyc Corp. All natural ontology
would be formative and thus express with real time alignment accounting for new
facts and allowing human intervention.
The formative process would use substructural ontology and a form of
qualitative structural activity relationship (Q-SAR ) analysis.
The resources for producing the
formative process can be standardized. Like in physical chemistry, the nature of regularity of these
resources (atoms and compositional rules) reflects the regularity of the social
discourse. If something new occurs,
then this something new is represented within the set of atoms and
compositional rules.
(for more on this discussion see: Tread on the
National Debate: [64] [65] [66] [67] [68] [69] [70] [71] [72] [73] [74] [75] [76] [77] [78] [79] [80][81] [82] [83] [84] [85] [86] [87] [88] )
The term “taxonomy” is properly used to be a set of terms with some
meaning and relationships indicated.
Most often the relationships are all subsumption relationships, but one
can attach other relationships such as similarity and dissimilarity. Word net is often used as a thesaurus. But again, the problem is often the closure
that occurs if a computer program uses a definition of the meaning of a word
without some real time reification process.
This reification process is why one needs to see topic maps as having something significantly different than
RDF ontologies.
The reality is, I'm not an ontologist or any other type of expert with enough knowledge to understand what was in it. Besides, I'm limited in my job and personal time to focus on one narrow slice (a common upper ontology).
If you think the Army shouldn't adopt a common upper ontology, you can do so by presenting a logical written argument why. I have posted this same question to multiple forums. Many are against it, but non have given any good reasons.
I believe one of your points is that computers can't understand or have knowledge.
Technically you are correct, but when people use these words, they mean a computer can process data and inference with it. When I started typing this sentence, I could say the computer understood I meant to type a capital 'W.' OK, so it's a computer and can't understand anything, but it did process my key strokes and produce what I understood to be my intent.
Information systems can't understand data, but with a common ontology, they can process it much better. How is this nonsense?
The nonsense is the attribution of human like cognitive properties to
ontology with first order logics. One
could model the known science and develop artificial neural network
architectures, and then one can get very interesting results. This is done in various communities, but has
not solved any of the problems with formal closure. This current failure does not mean that significantly more
useful machine computational architectures cannot be developed to
address the representation of human knowledge in real time. But even then, if one is not careful, the
human metaphor should not be used in a way that misleads policy leaders and
others.
Today, the Army spends millions mapping data between different systems, so the systems can utilize it and produce quality outputs. This works today, so how is this nonsense. But it gets too expensive with each new system you want to access, hence the wisdom of common data models, or better yet, ontologies. Tell me where this argument is nonsense.
The work on data integration and mapping is costing a great deal, and
it is often found to be partially or completely unsuccessful. On the other hand there are a number of
approaches that are not considered.
Even in cases where standards are produced, such as human mark up or
topic maps, if the political work is not constantly participated, then the
contracts go to those who spend the time and resources of courting the program
managers.
I hope that the position that my group is taking is clear. The issues are more complex than what the
question “should there be a single upper ontology for the Army”. This complexity is not realized by those who
make funding decisions, largely because of the marketing of simpler approaches.
The more complex approach has to deal with an absence of the type of
education that is needed to even have discourse about what the issues are. While at the same time, simpler approaches
that have little chance of doing anything.
The National Project to develop a K-12 curriculum in the knowledge
sciences is thus the required solution.