Thursday, May 19, 2005
The next generation knowledge tools
Two communications from Dr Richard Ballard, Founder of
Knowledge Foundations, with foot notes
by Paul Prueitt, and links to additional comments that might be made by
BCNGroup members
[79] ß First communication
Paul & Friends:
A conceptual view of a combined N=7&8 n-ary bundle is represented within a Mark 3 semantic web . Some examples can be given.
These theory bundles represent all possible theory-options and situational alternatives within a declarative theory based semantic web. N=7 & 8 indicates that the different reasoning paths available involve 7 or 8 degrees of freedom. In contrast to algorithms and linguistic description, semantic web pathways are unconstrained and free of context walls and barriers that force the reasoning path to start at one particular step and flow in just one direction.
The n-aries n-degrees of freedom contain sufficient conditionals so their application is possible in any sequence of application and so are everywhere reversible (any m conditions can be treated as independent and all known predictions for the remaining n-m can be treated dependent.
This property makes every bundle information conserving, when reasoning in any direction. This is equivalent to the thermodynamic (2nd Law) condition that there is no entropy change (information is conserved) if a process is reversible (maps independent<->dependent). The erroneous assumptions have been that communication is information conserving, but necessarily computation is not.
Side Note: The fact that
information is conserved in matrix inversion was first demonstrated by me
(Ballard) in a 1976 NSF study of quantum mechanics simulations. It established
that the explosive complexity was a consequence of mathematical correctness
possessing no constraints of parsimony (efficiency measure -- Ockham's Razor),
ie computational complexity is not a natural condition, it is a consequence of
choosing the wrong system of representation. Two different algorithms may both
be correct, but one may be non-computably inefficient.
Knowledge science virtually guarantees linear (proportional to information content) computing costs by separating theory (a' priori) computing costs from information computing costs (a' posteriori).
The declarative semantic web is a consequence of precomputing all theoretical constraints just once in all history, when the theory is learned. Those costs need never be repeated, today's computers pay them over and over endlessly every time the same question is asked again of a brainless machine with no memory that it was asked and answered millions of times before.
The useful life of theories
so far have been 2 million (human social interaction with more than 2000
individuals to 40,000 years for organization of business practice and formation
of states involving 50,000 or more non-related strangers. The newest possible
theories require S-T cycles of 30-50 years because it takes that long to create
and teach the new word meanings used. [1]
In contrast to theory, information constraints are applied when any real or hypothetical situation is made apparent. Its natural physical costs are measured then in physical bandwidth consumed when supplying information and decisive choices, this is exactly equal to Shannon's measure of information content. This provides proof that given a paid-for theory-based semantic web, computing costs should never be worse than proportional to information content. The fact that most costs today are quadratic to NP complete can be squarely blamed on uses of logical and mathematical representations that are not well matched to the processes they are used to model. Natural laws contain already principles of parsimony in conservation laws and least action. [2]
These concepts were organized into the draft paper called "Information, Structure, and Inference -- A Physical Theory of Knowledge and Computing" -- in 1993 and reviewed with the earlier matrix findings. The reviewers with Digital's AI unit confirmed their importance, but recommended these be patented and not published. Both were first announced publicly in 1998 and very widely discussed on Paul's forum and that of KMCI. [3] This theory sets bit limits for measuring the impact of any decision based upon theory. These measures tell us human knowledge is probably more than 90% theory.
For business models, information produces service organizations delivering the news. Theory with its natural life-times measured in thousands to 10s of thousands of years is the rational equivalent of real estate. Humans come and go in less than a century, but theory retains and grows in value for 10s of thousands of years.
The semantic maps that I have presented only begin to tell a revolutionary story.
Dick
[1] I find this type of fact to be useless because it is a statement about an abstraction. Clearly the eureka experience is the formation of a theory in an instance. What I want to know is the evolutionary psychology or the cognitive neuroscience that gives me true knowledge about the processes involved in creating new theories. I am also interested in who “owns” the theories and if this ownership issue will be part of our future.
[2] Again, I find fault with this paragraph in that (1) I have heard this language (about Shannon) before and many people have not, so there is little basis for communication; and (2) there are all kinds of ontological commitments made that I would like to bring objective light to.
[3] Ah, the good old days.