AN ADAPTIVE, EVOLVING NETWORK
[MAVIN: Machine Adam Virtual Information Node]
“We are not stuff that abides, but patterns that perpetuate themselves.”
Norbert Weiner, The Human Use of Human Beings
The following is a concept for creation of an adaptive, evolving network mimicking biologic processes. The network would rely on a simple organizing principle from which complex behavior and organization would emerge. The basic organizing principle upon which the concept relies is the attainment and maintenance of system stability, stability being envisioned as a basic organizing principle of both living and non-living systems. For programming purposes, stability is to be defined in terms of the number of relationship patterns between node and system data elements. The network is envisioned to consist of multiple interconnected nodes, each node to be programmed with a threshold level of stability which it must maintain. Individual nodes exist in a connectivity relation with their "data field," or areas of sensor data input, as well as other nodes. Inter-node communication consists of feedback and feed-forward loops by which inter-nodal pattern matching information is exchanged and attainment or loss of individual node stability messages are sent to other nodes. Upon reaching the threshold level of instability (number of unrelated data elements in relation to the existing node data base), individual nodes send an "instability" message to all other nodes to which they are connected. Upon receiving the instability message, external nodes are connected to the unstable node’s data base and begin to query their own data bases in search of data relations between their data and the unstable nodes data. As these relations are found, related data elements are forwarded to the unstable node until it reaches a predefined stability threshold. Upon re-attainment of system stability, the originally unstable node then sends a second signal to the other nodes with which it is connected to signal them to recommence normal internally stabilizing functions. The central node stores inter-nodal information relationship patterns to create a field system mapping.
Threshold levels of stability or instability may be defined in terms of a percentage of the number of relations to the total data field size. Threshold levels of stability assignments per node might be defined in relation to the importance of the individual node in overall system stability, the system being defined as the sum total of all of the individual nodes and their data bases (total system relations attained or determined/total system data). A central node would control system stability, selectively shutting down non-essential data nodes depending on the level of system instability. The central node would also store holistic system patterns and contain a mapping or addressing from pattern memory to contributing nodes. Lower threshold signal strengths would be required for inputs from nodes having significant contributions to overall system patterns, signal strength being adjusted based upon the degree of “particle resonance” exhibited by outputs from outlier nodes to the central node (relationships formed by inter-nodal pattern matching). The initial state of the central node would be set at a normalized required signal strength across all outlier nodes, threshold signal strength addressing being developed as outlier nodes compete, and this competition being category specific so that a mapping from the central node to outlier nodes develops specific to information pattern categories.
The following definitions of basic terms are provided.
Basic system construct = system instability elicits node/system actions to achieve the system goal.
System Goal = stability
Stability = information relationships
Relationships = repetitive coincident events or patterns
Positive Value = input or activity increases overall system stability
Negative Value = input or activity decreases overall system stability
Instability = lower threshold of repetitive coincident events, patterns/total data elements (system and/or node)
When the threshold level of instability is reached, the system enters a state of "heightened vigilance" in search of relations between information elements. This heightened vigilance may incorporate several tasks;
1. Search of system data to increase node data relatedness/identify previously unrecognized relationships or relationship patterns.
2. Passive or Active Selective Attention to sensory data with a higher probability of success in finding relations.
- Passive Selective Attention – no sensor node/sensor data field interaction
- Active Selective Attention – may entail the need for the data system to position the sensor to better advantage within the respective data field (change the degree of view of that which is being observed).
3. Active Searching – involves actions such as probing the data field or other actions wherein the sensor system searches for elicited responses to system actions.
4. Functions previously created through the observed data, and subsequently turned on that data using it as substrate, rearrange protodata into new relations (the system acquires not only new facts (datum), but also new ways of looking at those facts – what emerges can affect what it emerges from). The functions created or derived from previously observed repetitive coincident events can act as a type of spell checker on expectation and reality. When previously repetitive coincident events no longer coincide, as determined by an error between the new data, the old function based on proto-data, and reality, the function becomes hyperactive (hyper-vigilant), triggering the search for emergent relations and, as they are found, the development of new functions based on new data relations, and the revaluation of the old functions.
Data relationship definitions would be based on biological data structural relationships as depicted in figure 1 below. Specified triplet numerical sequences would relate to specific letters/codons, letter sequences to words/symbols, word sequences to phrases/strings, phrase combinations to concepts, and concept combinations to functions. Initial/seed individual data node inputs could run the data spectrum from simple quantitative data to functional data, the quantitative data being a representation of an underlying function (preformed relationship in the data field). No pre-established database would be provided. Rather, the system would learn through recognizing relationships within each nodes own experience as well as the experience of other nodes and sensor fields, applying this "understanding" to create additional relationships to further increase stability. The discovered complimentarity of information from other nodes provides a new informational figure/pattern from which to push and stack from within the individual node data structure, as well as a new epi-informational function to express lower level relationships (environmental feedback triggers expression of previously dormant information by providing new context for conceptual nearness of lower level symbols/patterns).
Douglas R. Hofstadter makes this same comparison, offering an analogy to biological system transcription, translation, protein manufacture and function using Typographical Number Theory and the Gödel Code in his book Gödel, Escher, Bach, figure 100, page 535, describing a Gödel Code for a similar process. Figure 2 below depicts his number to codon matrix which could serve as the basis for creating the mathematical genetic code of the network.
Biologic and physical system analogies can be carried further as follows. Biologically, each node could be representative of a specialized cell type, the inter-nodal messages to be hormonal messengers (a substance [relation] produced in one part of the body [data system] having an impact on another part of the body [data system]), and the functions produced by the relationship patterns to be enzymes. The last, the functions produced by the underlying relationships, could be such that they are active functions, as they are created being run or acting on their data "substrate," that substrate being a particular data sequence within the node. Random variation in the underlying genetic code would produce function variation which would be selected for continued expression based on its ability to efficiently and effectively identify relationships in the data field. The population of nodes would compete as overall system stability is a system goal, the central node selectively shutting down those non-essential nodes creating system instability, the individual nodes shutting down functions ineffective in creating node stability. The collective nodal processes create the epiphenomenon of a network “mind” which acts on the nodal processes in a feedback loop to change them through the manipulation of information based on knowledge obtained from and driven by the utility function. In this way, the mind moves matter, directing the characteristics, activities, and capabilities of the network, the associative nodal state being akin to a mental state. Physically, what is entangled is not therefore primarily matter but rather information, the physical being an epiphenomenon or phenotype of the information and, more, an epiphenomenon which interacts with its genotype through complex feedback loops. The evolution of information entanglement throughout the system becomes the “hidden variable” which describes the probabilities of relationship identification and development within and between nodes. The evolution of the wave function becomes the evolution of information entanglement throughout the system. The choice of which question to pose to nature becomes the “patterns in the brain mirroring the brain’s mirroring of the world.” (GEB) Nature’s statistical choice of which answer to give becomes the system’s choice of its most stable pattern given the universe of possible relationships to form and given its utility function.
3 Dimensional arrangement of the information based on all
information inputs at time T1. The
blue represents potential new information in the input pattern obtained by
comparison with node/system data patterns.
Such new information could be an identical pattern or isomorphic as
3 Dimensional arrangement of the information based on all information inputs at time T1. The blue represents potential new information in the input pattern obtained by comparison with node/system data patterns. Such new information could be an identical pattern or isomorphic as shown.
Concepts (3 Protein Strands)
(3 Protein Strands)
● ● ● ● ●
Time Tn Space Sn ●
The total system would consist of multiple sensor data fields as figures 3 and 4, each field connected with the others via a common interface for data exchange, the inter-nodal instability messages applying both internally between sensor data field nodes as well as externally between sensor data fields. Figure 5 below depicts one sensor field set up using the ART example, the central node here being equivalent to the F2 node in the ART system storing that fields long term memory, a system wide central node with this schema being expanded to the system level being envisioned. An example of a relationship pattern would be the repetitive appearance of a particular sensory data input with another sensory data input, some threshold level needing to be determined to separate causal from chance occurrences (noise) of coincident inputs. Within an individual node, and prior to input of relationship data from external nodes, a relationship pattern would be the repetitive coincident appearance of data from sensory nodes (for instance, the sound of a train whistle coincident with the sound of the wheels running on the tracks). Repetitive coincident information from different sensor types or from different inputs of one sensor type would serve to bind the information, similarly to the wiring together of neurons which fire together. Data elements would be time and location stamped or tagged to determine inter-nodal correlations of sensory data.
Figure 5 – Sensor Field 1 Standing wave (second sensor field)
The system mapping of the information patterns from the sensory field inputs becomes a function (a unique mapping from a domain space x (sensor fields) into a range space y (the system information topology). Transcription of input data sequences and storage in memory would create an isomorphic data set of the input information. Random variation in the transcribed input sequence can generate new potential mappings to increase relatedness (stability). Such variations would be compared to inputs across the system to determine coincidence with field inputs (do these patterns which could increase my stability exist somewhere in domain space x – the real world?) Such activity would be akin to the system thinking about itself as opposed to simply recording stimulations from the environment. If one considers the 3D arrangement of information as in Figure 4 an information particle, its spatial coordinates are its mapping to its local field map and its higher system map. Specifying all of its coordinates corresponds to specifying its spatial relationships to the entire range space y – node and system range space. There exists a mapping or activation probability across the range of potential linkages (vertices and edges) within the system topology and multiple potential particle locations which correspond to particular activation patterns. As information relationships increase stability which decreases energy, we can equate the total potential activation patterns to an effective potential energy surface and the specific patterns to potential energy wells, or the regions surrounding the local minima. As in Adaptive Resonance Theory (ART) systems, resonant behavior is equated to information relationships. The wave function describes the probabilities of relationship development within and between nodes. Oscillatory behavior corresponds to the particle being transiently trapped in a potential energy well (a local minima). Alternative minima exist throughout the range space. Oscillations are produced by the system transitioning to innumerable potential information relationships, the system seeking its most stable energy state. That state is equivalent to the greatest number of edges between vertices (relationships) throughout the system at any particular time. The closer the system is to the threshold value of instability, the higher the oscillation frequency as the system becomes hyper-vigilant in seeking to lower its energy state. The information “particles” from an individual sensor field combine to form an information “atom” as shown in figure 5 above. (As an alternative view to physical nodal connection, and as a concept linking pure information to matter, consider the following. These “atoms” resonant behavior produce outgoing oscillations/waves which interfere with incoming oscillations from other sensor fields producing wave interference patterns which characterize each sensor field’s relationship to other fields. These patterns are what are captured by the R/EM sensors. Stability, in this sense, is achieved through the creation of standing waves as the “atom’s” oscillatory wave, produced by its internal relationship patterns, relates to/interferes or resonates with incoming waves from the system. The resulting standing wave creates shells of matter as in the standing wave theory of matter.)
The following quote was written after reading portions of GEB. “The feedback loops would create sentient systems, the individual unitary elements of the system possessing no sense of self or direction with a purpose. It is only in association with other elements that the laws governing their social existence can be felt in their impact and reacted to. It is through the numerous, countless feedback loops that individual elements, as part of a complex system, sense their relation to that system and themselves as a unitary part of it.” Stability equates to harmony which equates to a minimum energy state. The harmony is achieved through known relationship patterns with both the internal and external environment. Disruptions in those environments create disharmony, or a disruption in the relatedness of the system parts and therefore a higher energy level due to the active search for new relationship patterns due to the prime system directive.
The following ideas were written after reading “The Intelligent Universe” by James Gardner. “The utility function = information relationships. Nature selects that which is in right relationship with its environment. That is in right relationship with its environment which 1) through random variation or 2) through knowledge is so ordered or has so ordered itself or its environment to achieve stability. Random variation and natural selection create the nascent forms which gradually acquire the ability to so order themselves to achieve stability – adaptability – and eventually the ability to control and even create their environments. The falsifiable proposition – that intelligence and sentience will evolve.”
“ ‘In the beginning was the Word, and the Word was with God and the Word was God.’ The Word – information with a purpose – knowledge. Purpose – to create life and, in particular, a life which could have a relationship with its creator – intelligent life – and through doing so with His creation. Perhaps we must evolve beyond the constraints of our biological intelligence in order to actualize our innate potentiality to know God through understanding the intricacies of our relationship with His entire creation.”