Monday,
April 11, 2005
Global
Information Framework
Hypothesis formation in the Trilevel architecture
Paul:
An additional question - where in your formalism does hypothesis formation take place and how is hypothesis uncertainty represented and updated ?
Regards,
W.
The question goes to the heart of why differential and formative ontology is quite different from artificial intelligence. We can follow the soviet applied semiotics school as discussed at:
http://www.bcngroup.org/area3/pprueitt/kmbook/Chapter6.htm
To do this requires a paradigm shift to a “stratified architecture” where invariance across many event measurements produces a set of semantic primitives expressed as elementary patterns measured using co-occurrence.
The event Structure Ontology Framework (e-SOF) is one way to constrain the measurement output into a regular structure, where this regular structure will produce (it is claimed) a type of periodic table that is associated with a specific complex process.
Unlike the chemical periodic table with the physical science “atoms”, the system specific periodic tables are “quasi” stable in exactly the way one would expect (for example as the predictive or anticipatory framework to ontology about a terrorism cell, or a importer of trade commodities.)
!! This conjecture about a quasi-stable “periodic table” underlying the formative causation of any complex system is the key element to the notational foundation to the stratified theory. Artificial intelligence has always been aligned to a non-stratified and reductionist viewpoint where things like expert systems are treated with greater respect than things like connectionist viewpoints (evolutionary computing, neural networks, latent semantic indexing).
Connectionist theories, do not quite make it to a fully “stratified theory” as I have discussed in one of my chapters. Connectionism has top down constraints, but not often a theory of semantic primitives. Counter-examples are the work on explanatory coherence by Paul Thagard and harmony theory by Paul Smolensky. But these works have not been extended beyond their original statements.
The proposition that OntologyStream Inc has put together, (the six page deployment proposal authored in early April 2005), is that a composite semantic extraction technology should involve some easily manipulated measurement process, with an action, cognition, perception cycle something like the Actionable Intelligence Process Model, as discussed in the PowerPoints on e-SOF and Differential Ontology Framework (DOF).
The technology composite should reflect, and compensate for, all relevant enforceable patents on underlying processes.
The planning and design of the Global Information Framework (GIF) seeks to find a business process that supports, and does not destroy, an integration and extension of:
1)
The
best semantic extraction processes, developed with deep understanding of the
underlying patents to five or six COTS products, with full compensation to
these intellectual properties.
2)
The
extension of OWL, and Topic Map, type representation over sets of concepts to a
form of compact representation of sets of concepts using a key-less hash table
data encoding mechanism (that is provable optimal in terms of the use of
convolution operators).
The two keys is computability and visualization of instance scoped Ontology Individuals (is-OIs). Computability is provided by the DOF (via the two above mentioned extensions.).
Neither OWL nor Topic Maps have nailed the ontology visualization issue. This issue is aligned with cognitive neuroscience via the use of the subsetting and a perceptual physics based interface design.
This design allows the is-OIs to form in real time under the complete control of a perceptual element to the formation of the is-OI. Instance scoping can be mediated by the work I did on extending the quasi axiomatic theory foundation to Pospelov and Finn’s Soviet area applied semiotics, as discussed in Chapter Two of the Foundations.
The development of instance-scoped ontology, as a differential and formative process, is HIP (Human-centric Information Production).
The underlying ontology persistence is now based on Orbs (my invention that is perhaps slightly infringing on the Hilbert Engine patent (2004) by Primentia) and the convolution operators defined as an extension of the Applied Technical System (NdCore) patents (1991, 1993) on “conceptual roll-up. Both of these are related to the earlier Readware patents (1986) on stem based n-gram measurement using an internal ontology. Recommind patents (2004) and IP from SchemaLogic begin to round out one set of IP that could be integrated as a technology composite. There are a few other ingredients, which I will not mention here.
In the near future, we hope to see these key patents reexpressed in Cubicon, as Klausner gets CoreTalk corporate structure in place. One has to say something about the evolution of the IP space, and the principle of sufficiency since we are well aware that some patent owners will not allow the technology composite to include or infringe. I can suggest a few whom I think will not participate. However, a bit of little known mathematics (developed by Harold Szu, and discussed in important circles within the neural network community (1998 – 2000)) creates flexibility to go around, through and to set aside patent ownership where intractable positions develop. This issue goes to the “business model” expressed in 1997 in the BCNGroup Charter.
The important thing to realize is that the kind of IT procurement now apparently required by law inhibits the formation of a process that could bring differential and formative ontology frameworks into the market. Thus we identify the requirement to act in a political fashion.
The BCNGroup Founding Committee predicts that market forces will come to our aid at some point. Why? The answer is value to society. The literal value to society of this composite technology is not measurable. The opposing value, in not having this composite technology in the market place, is value (estimated to be at least 30 B per year in direct, and wasteful, time and materials contracts) to the current IT vendors and consulting firms.
Moreover, the IP space has not even been defined, except for the early patents, like those I mentioned, and others like those of Wycal (Artificial Linguistics and Oracle ConText).
See also the discussion about Pribram’s holonomic theory about perception and cognition. (In several places in my on-line book.)