Friday, January 27, 2006
[145] ß [parallel discussion on
generative methodology
Concept representations without deductive logics
On the face of it, it is possible to assert that natural systems, ie non-formal system, do not exhibit deduction. It is also possible to assert that no natural system is fully formalizable (Rosen, Penrose, etc). Obviously one places "formal processes", as executed by a computer, is a non-natural system. But this is something that is done carefully.
In this viewpoint, we admit utility to mathematical models and logical models in certain circumstances... but not in all circumstances. The BioPAX working group is considering two objectives
1) integrating massive relational data bases (SQL) into a single
sharable information system based on the use of OWL Full.
2) assisting natural scientists to conduct investigations about the nature and function of cell and gene expression.
In several current standards and working groups, there are serious discussions about forming an ontology in a two step process.
The first process is to define a set of concepts ... OASIS SOA IM work is a good example. These concepts have a name and "internal data definitions" and thus look very much like objects in the object oriented paradigm. I attach the draft SOA IM so that others can see precisely what is involved.
The second process is to consider the question of retrieval of information.
These discussions are deeply informed by lessons learned from attempting to develop operational ontology for computers based on an incomplete understanding of the nature of logic, the nature of the world, and the relationship between the two.
In so many cases, what the biologist thinks she/he is getting is not at all what the computer will actually retrieve. BioPAX has gone a long ways in determining what it is that the biologist wants. By using OWL Full in a sometimes non-standard way, they are getting some of this.
But in each case, one gets retrieval of information due to some specific programmatic functions, not inference; where
inference sameAs human inference
I would be interested in a example where "inference" is not purely retrieval. Dr Obrst, do you have such an example? Dr Finn?
The issue is not merely that OWL logics are retrieval mechanisms but that the full reality of human inference is very poorly represented by deduction.
Plausible inference using computational means is not attempted because one finds few cases where the classical knowledge engineers get away from the assertions that set up deduction as a structural mechanism. By plausible inference I mean the work by the Soviet school, not the probabilistic theory called "plausible reasoning in the West" (use google). The Soviet school used Mill's logical cannons.
I know that my notes here are not popular in many circles. But the "Second School" viewpoint is grounded in scientific literatures and can address any issue posed by the "First School" viewpoint. The historical fact is that the Second School has never been allowed the kind of funding that the First School has received; and thus has not developed a cadre of 1,000 or more PhDs to act uniformly in peer review etc.
The point is that the type of deductive mechanism developed in OWL Full (or other "expressive" logic) is merely a complicated retrieval mechanism. There is no magic; you get what you put in and only what you put in. Nothing more.
With human reasoning there is magic, and the nature of this magic is not well understood.
This does not mean that computers supported inductive reasoning is not possible; but here one has to understand the nature of human reasoning and not make
human reasoning "sameAs" computer function
As Alan remarked, correctly, not long ago; the purpose of BioPAX is to move bioinformatics information from relational databases, where the schema are very brittle and inflexible, to XML information bases with organizing schema based on the RDF model. RDF is used to express class - sub-class hierarchies, and these serve in a classical "taxonomy" fashion to organize how data is stored and retrieved. RDF used in this fashion blocks off the development of n-ary representations, such as what Drs Krieg and Ballard (and others) have developed:
http://www.bcngroup.org/area2/KSF/Notation/notation.htm
The RDF/OWL evolution is thus seen as a evolutionary dead end, but full of utility as a step beyond the relational (SQL) database. The more that is invested in this RDF/OWL evolutionary path, the harder it will become to get back onto a pathway towards true human computer interactions based on an understanding of respective natures.
Dr Paul Prueitt
The Taos Research Institute
Taos New Mexico