Hitting the wall
To date, most approaches to resolving information overload in the face of knowledge-driven objectives has fallen into a prescriptive structured approach, where the data is organized ahead of time in ways presumed to meet future need. We see this approach as prevalent in structured data-warehousing where information must be extracted, aggregated, and transformed into fixed summary structures to be responsive and scalable to business needs, and in unstructured data where neural-nets and Baysean Belief Trees are compute-expensive and time consuming to generate. These methods have proven too complex and unresponsive to use and maintain in today's world where structured systems typically get left behind in a sea of change.

Any requisite for manual intervention, typically required with structured and declarative approaches, creates a manual labor dependency that cannot be sustained, particularly in the face of depleted human resources in today's business challenging economy when budgets must be deployed towards customer facing opportunities, rather than internal infrastructures. That's why we see such a demand for CRM (Customer Relationship Management) systems and for maximizing resources deployed by leveraging "what we know" through eCollaboration Systems. 2G-KM technology can help us capture and reuse effective knowledge concept patterns contained in documents that litter our networks in order to meet this challenge.

Leveraging concept relationships in unstructured documents
Emerging 2G-KM technology can automatically normalize the language used in documents into concept hierarchies that emulate directed graphs, as shown in Figure A, tagging hierarchy nodes with common vocabulary terms, from which it can build concept patterns called Knowledge Signatures (kSigs).

FIGURE A


Here is a partial document knowledge signature. Roll over picture for a larger image.

Document text can be componetized around concept nodes at the paragraph and phrase participation level. Paragraphs can participate in as many nodes as appear in their text, creating an automatic concept relationship topology that cross-associates as a latticework through common context.

A new indexing fabric: everything interconnected
Indexing a latticework that can necessarily form an ordered tree of arbitrary depth from any perspective, emulating abductive inferencing for real-time machine learning, and supporting inductive and deductive inferencing for causal reasoning, is a tall order for today's standard indexing methods. Such an indexing system must necessarily connect everything to everything dynamically in responsive fashion across massive volumes in a way that's not compute-expensive, so that it can operate on standard computing architectures.

Intelligent Knowledgespace, a key ingredient
New 2G-KM multi-dimensional associative indexing methods are required, such as fractal indexing in high-dimensional space, where dimensions become dynamic attributes. In such an index, dimensions can not only be dynamically reordered from any perspective on a sub-index basis with real-time response emulating PointCare sections, but they can also be reflexively self-aware so that the data itself can actively read and reordered along optimal paths based on ambient multidimensional context influences and can draw meaningful inferences that bubble-up to applications at the surface through a method of dimensional summation (perhaps generating Hamiltonians).