Transcript Rationale of the Paradigm

Orion
Philosophy and Rationale
“Knowledge Is Structure”
If it really is structure,
what sort of structure is it?
We are asserting that it is
active dynamic undirected structure
A Simple Structure
The structure can propagate a
wide range of entities through its
connections, and the operators
can operate on analytic or
experiential information, and the
structure can change itself.
• Visible
• Undirected
• Dynamically Extensible
• Controllable Existence
Logical Surface
Statements are written on a logical surface
a=b+c
Building Structures
We can combine
small logical
structures into
larger structures
without concern
for how they will
later be used.
The directions of
information flow in
the structure are
dynamic
IF a + b = c THEN d + e = f
Argument from Similarity
Real Neurons (not artificial ones)
• Switching
• Complex messages
• Back connections
• Changing connections
• Storing of memories
• Massively parallel activity
Object
Network Operators
• Switching
• Complex messages
• Undirected
• Changing connections
• Storing of memories
• Quasi-parallel activity
Time
Object
Time
<
But Real Neurons Are Directed
How can the fluidity of knowledge be
built on directed structure?
Back-connections throughout the
neuronal structure can hide this
physical directionality operating at a
lower level, just as many small tasks in
a queue can hide most of the
limitations of a single process
Argument from Function
A mental model has to represent a fast-changing world.
Its attributes are no accident. The connections in different
areas may be different, but the substrate is universal.
A knowledge model has to represent a changing world.
Can its substrate also be universal, so there are no
boundaries between different aspects?
Argument from Utility - One
An algorithm/data structure has
one part, the data structure, that is
visible and easy to change, one
part, the algorithm, that is invisible
to the machine and hard to change
- a stream of instructions.
An active structure is all of it visible
all of the time. Changing the
structure changes its behaviour.
Argument from Utility - Two
One algorithm/data structure
does not easily merge with
another - each has been built for
a purpose.
One active structure easily
merges with another - the
phasing lies in the structure, so
connecting new structure
changes the phasing to suit.
Argument from Visibility
Procedures on a stack are invisible to everything but
the stack handler. Nothing else can see what is
happening and respond to it - the machine is blind to
its own workings. This prevents quasi-parallelism.
States being propagated in a network are always
visible - anything that can be connected to a point of
interest can respond to change or influence it, down
to the limit of operator granularity.
Argument from Dynamism
A=B+C
Static structure
X = SUM(List)
Dynamic structure
The quake struck early Thursday.
Ultimate dynamism
All knowledge has dynamic aspects - aspects we can’t avoid.
If a paradigm can handle the most dynamic, it can handle the
least dynamic or the static on its ear.
Nowhere to Start
This is typical - knowledge is used everywhere no start, no end, lots of loops
- it is definitely not a DAG
Attempts to enforce some predetermined
structure quickly result in irrelevance
Orion Principles
• Destroy no inference in the transformation from text to
structure
• Discover universal substrate by continual extension
of problem space
• The activity is only in the structure, not outside it
• Invisible activity (within operators) is minimised by fine
granularity
• Accrete the structure where simple operators are
inadequate
Examples
Project Management
Development projects have a life of their own.
There is uncertainty in what you are doing, and whether and
when you will be doing it.
Critical Path Method doesn’t allow you to plan what to do, only
when to do it - the activities in a knowledge model have both
logical and existential control, and the design model can sit in the
same model as the activities and resources, all made out of the
same stuff and communicating with each other
Reinsurance Simulation
Household
The structure generated for Commercial
Property (each column represents a policy)
Generated Commercial
Property Claims
A Surplus treaty
Claims paid after the
Surplus recoveries
Risk Excess treaty
Claims paid after the
Surplus and RiskXs
Combine with Household
Claims after Quota Share
(into Catastrophe treaty)
The structure of the reinsurance program is in the form that insurance people
understand it – claims flow into a reinsurance treaty, which recovers some of
the payments. What remains can be combined and flow into another treaty.
The structure is easily modifiable and extensible because it is self-phasing
Combining Knowledge Domains
Airframe
Performance
Avionics
Assemble Pieces of Knowledge Into an Active Object
Which Itself Can be Assembled...
Extracting Meaning from Text
Section b(ii). An
occupies relation,
with Tenant as a
subject and
portion as an
object, and a
logical control
structure
controlling 2 pay
relations. Portion
is part (contained
in) Space, which
is the local
dictionary object
for Swing Space.
The occupy
relation gets its
TimePeriod from
the Extended
Swing Space
Term – the new
TimePeriod of the
occupy relation.
The StartDate of
both pay
relations is
01/09/2005.
Linking In Bioinformatics
Anatomy
ABCA1: ATP-binding cassette, sub-family A (ABC1), member 1
Genes
LocusID: 19
Overview
Family ABC (transporter across membranes)
Subfamily ABC1 (members ABC1, MDR/TAP, MRP, ALD, OABP, GCN20, White)
Gene ABCA1
Protein NP_005493
Substrate Cholesterol
Function cholesterol efflux pump associated lipid removal pathway
Mutation causes Tangier’s disease, familial high-density lipoprotein deficiency.
Liver
Brain
?
The membrane-associated protein encoded by this gene is a member of the
superfamily of ATP-binding cassette (ABC) transporters. ABC proteins transport
various molecules across extra- and intracellular membranes. ABC genes are
divided into seven distinct subfamilies (ABC1, MDR/TAP, MRP, ALD, OABP,
GCN20, White). This protein is a member of the ABC1 subfamily. With cholesterol
as its substrate, this protein functions as a cholesteral efflux pump in the cellular
lipid removal pathway. Mutations in this gene have been associated with Tangier's
disease and familial high-density lipoprotein deficiency.
Diseases
Kidney
Eye
Brain
Cortex
ABC
HKT
Proteins
Liver
PFIC
PKC
NP_0001
NP_0047
Chromosome: 9 mv Cytogenetic: 9q31.1 RefSeq
The structure is used to understand the text then the text is used to extend the structure
FF
How Practical Is The Approach
• Creation of model is fast if knowledge is available
• Incremental growth of model topology suits typical
knowledge acquisition cycle
• Large simulation models have similar execution
times to more simplistic programmatic versions
• Dynamic Constraint Reasoning models have similar
solving times to other more static methods
But Is It Easy
In one way, creating a knowledge
structure is an iron discipline - unlike
a program, effects can only occur
through connections - so connections
must be made and states propagated
- no lazy reaching out to assign a
value.
And yet, is there another way - how
can we be sure we are in the correct
state unless everything that is
relevant can see and contribute to the
current state.
Why Do This
As applications move towards the use of knowledge, the
limitations of a single point of control machine, with so little
in its focus, become increasingly burdensome.
Still limited to a single point, the propagation of states
through network connections, the continuous visibility, the
fine granularity and the self-modifiability increase the size
of the apparent focus and permit us to move closer to the
full use of knowledge.