Holarchy

Work in Progress: This article is still in the developmental stage. Please do not base any work on this content, any aspect of it may undergo change or be removed at any time without notice.

A holarchy is a hierarchical structure in which each part or component forms an autonomous whole that is also a part of a larger system. This concept was introduced by the American philosopher Arthur Koestler, who argued that complex systems could be understood as interconnected networks of holons – self-organising units that are both parts and wholes at the same time. In other words, a holarchy is a structure in which each part has its own identity and purpose while simultaneously being part of a larger whole. Koestler called the two aspects of a holon the self-assertive and the integrative behaviours respectively. He described the former as an inward-facing system operating with flexible strategies within an individual holon. The latter as an outward-facing system operating with fixed rules as a network, or holarchy.

We believe in the idea of a libre society in the same sense as libre software, and that organising as a holarchy achieves this. We, the people, must figure out for ourselves how to live and work together as a single organism, it's not in the nature of centralised governing powers to do this for us.

We're not trying to compete with or change the current entrenched mechanisms of society at large, but rather just create a network that we and our own close group of organisations and projects can use amongst ourselves. Once we've tested and refined it to a stage where we're finding it really useful ourselves, then others will naturally find it useful too. As it grows in this organic way, its utility as a whole will grow exponentially due to the network effect.

Etymology

The suffix "-archy" comes from the Greek "archein," meaning "to rule" or "to lead". Based on its etymology, "holarchy" implies a form of organisation or structure where each unit (or 'whole') is both a part and a whole in itself. It suggests a hierarchical system where each level or unit is a self-contained whole that fits into larger wholes. This concept often appears in systems theory and organisational studies, emphasising the nested, self-similar, or fractal nature of systems.

Another closely related word is "holocracy". Both concepts involve a recognition of "wholeness" in each part of a system. Holarchy integrates this with a hierarchical structure, offering a balance between rules and autonomy, whereas holocracy is more inclined towards ensuring autonomy and distributed governance.

The term "holarchy" better describes our system, because it's explicitly defining a hierarchical system of authoritative rules - collaborative rules that provably maximise individual autonomy. "Holocracy" best fits a purely decentralised system that does not define any concept of authority.

Technology and Web3

The fractal nature of life allows us to equate the biological cell with a person, and a single person with the planetary organism. In his book Spontaneous Evolution, Bruce Lipton shows us that we can learn from our cells how to live together in peace and harmony as a single organism since they're a living example of it, and have been doing it for millions of years.

Many people who are strong believers in the idea of humans living fully in accord with nature think that technology has no place in this vision. But by looking at how the cells in the human body are able to live together as a community with a population of over fifty trillion reveals that technology is essential. The cells manufacture and maintain huge infrastructures including the equivalent of buildings that are tens of thousands of stories high, sophisticated networking systems and even an energy based financial and banking system.

Holarchy offers us a framework that can achieve this kind of scale-independent organisation, and can be understood clearly in terms of our own technological infrastructure.

The internet connects all of Humanity, and is evolving into an ever more complex, resilient and organised system. It's organised in layers of open protocols from the most fundamental physical layer up to the high-level organised layer of high-level application protocols. In addition the internet is generally referred to as having gone through a few different versions or phases, the first was characterised by servers and tech specialise being responsible for generating and maintaining the content. The name "web 2.0" was given to the broad phase that cam with blog and wiki software in which the vast majority of content was being generated by the users. The meaning of "web3" was originally used to refer to the semantic web which was envisioned to be a new level of organisation of the web's content brought about by metadata annotations. But web3 started to slow in its progress with corporate interests gravitating instead towards deep learning and AI as solutions to organisation. The "web3" term ended up referring to the decentralised nature of the web which started gaining popularity with the introduction of blockchain technology.

The term "web3" has evolved in its meaning over time. Initially, it was associated with the concept of the Semantic Web, which was a vision for the future of the World Wide Web where information would be structured in a way that machines could understand, interpret, and process it more effectively. This would involve adding metadata and context to web content, making it more accessible and useful for both humans and machines.

However, in recent years, "web3" has taken on a new and different meaning. It has become closely associated with the idea of the decentralised web, blockchain technology, and cryptocurrencies. This shift in meaning is likely due to the growing interest and development in blockchain-based technologies and decentralised applications. In the context of the decentralised web, "web3" represents a vision where online interactions, data storage, and services are not controlled by centralized entities like tech giants or governments but are instead facilitated by blockchain networks and decentralised protocols.

Interestingly, the holarchy architecture actually fulfils both definitions of "web3", because it maintains and evolving ontology in which all content is organised, as well as being able to function ideally in a fully decentralised environment. A unified ontology of knowledge as well as a map of usage of the ontology.

Our current internet protocol stack lacks a layer dedicated to coherent knowledge sharing and organisation. Currently these aspects are provided by a variety of specific applications. Since knowledge-sharing and organisation are so essential to a harmonious society, we feel that they should be provided at the level of the common networking protocols. A holarchy is just such a networking protocol, it allows participants of the network (holons) to interact together with a common means of organising attention and resources and of sharing, using and assessing knowledge. Holarchy is the organising principle and network architecture of nature and we believe, even of consciousness itself.

Our holarchy project

At Organic Design we're researching and developing a holarchy in the form of a p2p distributed network of holon-organisations, an aligned community of autonomous organisations.

The project's development effort can be broken into three general areas: the p2p network architecture, the holarchy organisational system and AI integration. In this article we offer a brief introduction to these three aspects.

The research aspect of our project extends out to a wider focus than the development to encompass the philosophical aspects of holarchy. On the network architecture side this wider focus is on political philosophy - the kinds of large-scale social order and progress that the holarchy system of organisation implies. And on the AI integration side, the philosophical focus is ontological (holarchy as a foundation for cognition and even of reality) and ethical (not just the AI alignment issue, but also the "human alignment issue").

• some examples of threads in our org that we can use throughout the text

Preliminary concepts

This section introduces some preliminary concepts that the rest of the material relates to, if the headings in this section are all quite familiar, then feel free to skip straight to the holarchy organisational system.

Free market

The free market dynamic, when actually free (i.e. transparent and unmanipulated), permits a harmonious balanced resource allocation system. The most fundamental systemic aspect of a biological organism is its bio-economic aspect. The market dynamic cannot achieve this harmonious potential alone, it needs to be balanced with the integrative behaviour of the shared ontology.

• in the context of organisation, ecosystem and exchange is needed
• knowledge from the edges

Object-oriented design

Object-Oriented programming (OO) is a paradigm that structures code around abstract entities called objects, encapsulating their data and behaviour. It promotes modularity, code reusability, and a more intuitive representation of real-world concepts, making it a fundamental and widely used approach in software development.

An object has a public interface and a private implementation. All interaction from the outside must occur via the public interface. The implemented within expresses some kind of agency that is responsible for presenting itself through the public interface. The agency can also perform services in its environment which involves organisation within. Examples of such agency are a program execution environment, an API, a large language model (LLM) or a user.

The main distinguishing factor between different types of OOP concerns how the system manages the "packaging" and deployment of commonly used functionality. We refer these packages "classes", and all OO is essentially a different functionality-packaging methodology. We use a class approach called "mixins" which is focused on composition of classes rather than inheritance.

The key takeaway at this point is that a class is an abstract package of usable knowledge.

• classes are shared (ontology) evolve (develop)
• OO describes autonomy (but some objects more than others, continuity, closure)

Software design patterns

Design patterns are tried-and-tested solutions to common software design problems. They represent best practices, not in terms of writing specific lines of code, but in terms of organising object interactions to achieve certain design goals. Think of them as templates or blueprints that can be adapted to individual needs, irrespective of the specific technology or language being used.

For example, consider a scenario where multiple parts of a software system need to be updated when an object's state changes. Instead of hard-coding each of these updates, the "Observer" design pattern suggests a model where objects can "subscribe" to another object's state and get notified of changes, ensuring a decoupled and efficient design.

The beauty of design patterns is that they provide a shared vocabulary for developers. When someone mentions a "Singleton" pattern, it instantly communicates the idea of a class that ensures only one instance of itself can be created, regardless of the specific implementation or language. By leveraging design patterns, developers can avoid reinventing the wheel, instead relying on proven solutions that enhance code modularity, readability, and maintainability.

Design patterns can be thought of as applying OO principles to the OO software development process. The design pattern paradigm within the context if OO can be seen as a reflective process of applying OO to itself, representing the OO ecosystem of patterns and projects as an instance itself. The is what the holarchy is, a kind of singleton instance structure undergoing evolutionary development from within.

The patterns operate hierarchically from the largest scale to the smallest. Local regions are best suited to know what more specific patterns are most appropriate in the own region. This is the same in the instance tree.

Production rules

Production rules provide a powerful means to represent systems and knowledge. A production rule consists of two essential parts: conditions (or antecedents) and actions (or consequents). These rules follow a simple "if-then" structure, where the "if" part specifies the conditions that need to be met, and the "then" part defines the actions to be taken when those conditions are satisfied.

Rules can be composed into complex work-flow structures, allowing for the expression of complex logical relationships. It is widely used in expert systems, business rules engines, and knowledge-based applications.

Production rules play an important role in automating decision-making processes, enabling systems to make reasoned choices, offer recommendations, and adapt to changing circumstances based on the knowledge encapsulated in these rules.

The control-flow is not referred to by the rules, rather it is entirely determined by the structure of rule composition.

The blackboard metaphor

Rule conditions and actions are all in terms of the representation in the local scope. The representation acts as a common abstract medium which allows the agents who actually carry out the interactions to be independent of each other.

• whole assures and maximises individual autonomy

This pattern of collaborative organisation is a popular software design pattern called the blackboard pattern which can be explained metaphorically explained as follows.

A group of specialists are seated in a room with a large blackboard. The specialists are working as a team to brainstorm a solution to a problem, using the blackboard as the workplace for cooperatively developing the solution. The session begins when the problem specifications are written onto the blackboard. The specialists all watch the blackboard, looking for an opportunity to apply their expertise to the developing solution. When someone writes something on the blackboard that allows another specialist to apply her expertise, she records her contribution on the blackboard, hopefully enabling other specialists to then apply their expertise. This process of adding contributions to the blackboard continues until the problem has been solved.

A blackboard system enables this flexible brainstorming style of interaction between diverse software specialists. Each of these specialists scans the changes to the blackboard, and posts an updated partial solution based on the state of the blackboard whenever its own internal conditions for doing so are met. These partial solutions cause other knowledge sources to update their portions of the solution on the blackboard until eventually an answer is found. In this fashion, the specialists work together to solve the problem.

The blackboard system is declarative which means all interaction is in terms of the shared space, the organisational methodology itself, such as workflow or queues, does not need to be discussed. The participants are decoupled from each other, and the organisational system is independent of participant schedules or implementations.

Continuous improvement

• needs updating for new position in intro

The nature of the four quadrant threads operating together also yields a continuous improvement loop like the PDCA loop. In the PDCA loop the four phases form a definite repeating cycle, whereas in our system the order is more abstract, because all the quadrants are continuous independent threads (all sharing the same instance scope as described above).

But in both cases, the result is an iterative pattern of continuous development, progress or improvement.

Adaptation is also specialisation, and the source of evolutionary change (although we could say the operation phase is the source since that's where the request emerges, but then maybe its the booking of that activity... etc...).

• iterative (standardisation/releases/adaptation)
• ratchet example of increasingly beneficial foundations
• adaptation is the key phase in the cycle (the creation of a new foundation locked in), which depends on virtual instantiation

Feedback loops

A feedback loop is a fundamental aspect of existence and change. It's a pattern that recurs in different forms across all scales and contexts, from the smallest quantum interactions to the largest cosmic processes.

Feedback loops are a fundamental component for building systems, we just refer to them as "loops" in this document.

Loops are constituted from two complimentary sides, an active side and a receptive side. Their specific meanings are context dependent, but the loop's sides are always complimentary. Some example are input and output, condition and action, stimulus and response, problem and solution. The loops are iterative with the active side making some kind of change, and the receptive side informing the next action.

The cybernetic loop

The cybernetic loop is one particular form of feedback loop. It represents a dynamic process where a system continuously monitors its output, compares it to a desired target state, and then adjusts its actions to minimise the difference, or error, between the two. This loop is also called an error-correction loop or negative feedback loop in some disciplines.

This iterative loop enables systems to self-regulate and maintain stability by making continuous adjustments based on incoming information, ensuring that they remain on course or adapt to changing conditions. From simple thermostat-controlled heating systems to complex AI algorithms, the cybernetic loop plays a n essential role in various fields, facilitating effective control, adaptation, and optimisation of processes and systems.

The cybernetic loop can be seen as a generalised continuous version of a production rule. Rather than a simple discrete action in response to an assessed condition, the condition is continuously monitored and incorporates the concept of a target state from which it differs. The action is continuously applied in feedback to reduce this difference.

Rule-sets are descriptions for agency that can act in accord with them to assess current state and work towards changing it to a target state. This is the cybernetic loop in action, but rule-sets can also be complex hierarchical structures of rules.

The pattern language described above and the agile style of project management are two examples of this same iterative cybernetic loop being employed at the level of complex real-world organisations.

The body schema

Another example of a complex hierarchical version of the cybernetic loop is the mental representation of our own bodies in our minds, called the body schema.

This internal representation and awareness that individuals have of their own bodies, includes their size, shape, position in space, and the relative positions of body parts. It plays a critical role in our ability to perceive and interact with the external world.

At its core, the body schema involves a continuous feedback loop where sensory information from the body, such as proprioception (awareness of body position) and tactile feedback, is constantly processed and compared to a mental representation of the body. This representation is adjusted based on the incoming sensory data to ensure an accurate perception of one's body and its relationship to the environment. This process can be hierarchical, involving multiple levels of abstraction, and it allows us to perform tasks with precision, adapt to changes in our body's state, and navigate the world effectively.

In essence, the body schema embodies a sophisticated form of the cybernetic loop. A holon has an information data structure that operates in this same pattern in accord with the cybernetic loop, but we refer to it in this context simply as the "representation".

The representation includes not only the current state, but also the future (objectives) and the past. The future is incorporated by acting as objectives for how the representation should be, the representation serves as an interface... it is an ontological representation of reality allowing it to be organised. Such representations of reality (the problem domain) are called declarative knowledge.

It's a lot easier to make the connection between the body-schema and the holon representation when we consider that our body-schema extends beyond our bodies in the form of tools and technology. And even beyond that into the wider culture and society as our values and property become part of our body-schema control structure.

Evolution

Evolution can be boiled down to an extremely simple dynamic in its general form. David Deutsch describes it as "the creation of knowledge through alternating variation and selection". The complexity we see in evolutionary systems is due to the evolutionary dynamic itself which tends towards ever more diversity and complexity. But the underlying dynamic responsible for all this complexity remains unchanged.

In reality knowledge is always evolving in diversity and complexity, because it's not just inert information, it's a dynamic process involving subjective values and application within diverse circumstances. The network protocol needs to facilitate this evolutionary knowledge process.

Culture is knowledge, knowledge is evolution, it depends on, builds on, and consists of, other knowledge therefore knowledge is always evolving in diversity and complexity. Knowledge and evolution go hand-in-hand, they're interdependent concepts.

Our genes, our culture, our society and our own minds are all structures of evolutionary knowledge, even though their media and selection mechanisms differ. Knowledge and evolution are interdependent concepts.

Memes

The concept of a meme was coined by Richard Dawkins in his 1976 book "The Selfish Gene". It refers to an idea, behaviour, or cultural element that spreads and replicates through imitation and cultural transmission. Just as genes carry biological information, memes carry cultural information, evolving and propagating as they're passed from one individual or generation to another. Memes can encompass a wide range of cultural phenomena, including customs, rituals, fashion trends, catchphrases, and more, playing a crucial role in the evolution of human culture and society. As we've seen in recent years, the internet has allowed memes to spread and evolve much more rapidly, and AI promises to multiply this still more.

Memes are a very similar concept to our idea of a representation (in the body-schema sense) within a holon which is effectively a "behaviour package" (a rule-set). Adaptation and evolution are enabled by all instances of the same class form a community which aggregates metadata about the packages and is automatically shared.

The network of all memes taken as a whole is the culture and is analogous to the ontology of all classes in our system.

This is the same as molecules, proteins and cells that make up an organism all being in flux around form determined by the organism's DNA. Likewise, our own mental cognitive symbols are in flux around forms within the collective unconscious.

Holarchy organisational system

A system is a collection of interrelated components that work together to achieve a specific objective. Systems can be physical, like a mechanical system, or abstract, like a software program. They consist of multiple elements that interact continuously in complex ways involving complex states.

The holon organisational system is a general pattern for organic behaviour that incorporates all the preliminary concepts discussed in the introduction. The system continuously improves, specialises and evolves. All scales of organisation working in harmony together as a coherent whole.

The holon system takes the form of a structured hierarchy of feedback loops very similar to the body-schema concept described in the introduction.

The two behaviours of the holon

The two behaviours of a holon

Koestler's self-assertive and integrative holon behaviours connect the public collective world beyond, and the private individual world within.

Note that this is a holon model, which means that when we say "collective" here, we mean the way holons behave in order to support a collective together.

Another note related to this is that it seems counter intuitive that the self-assertive behaviour comes from the collective and to the individual. But the movement from above to below represents a movement from external relationships and conditions of a holon to its internal intentions and actions, or in other words a natural movement from general to specific.

The two behaviours form the most fundamental relationship in the holon model, because it's existential in nature. Individuals and the whole (as well as "wholes" over all scales of operation) exist in an interdependent relationship supporting each other. The individual's primary requirement from the whole is a secure autonomy-condusive environment, which the whole assures it in return for its conformance with the protocol.

The interaction between self-assertive and integrative, individual and environment, is a feedback loop (a bidirectional relationship) in which both sides co-evolve together.

Co-evolution implies complexity which is structural depth and hierarchy. This is why the axis is drawn as a one-to-many tree in the diagram. Every holon is a hierarchical structure of holons, the self-assertive behaviour a flow of focus that divides within. The integrative results from the bottom-up integration of informative accounts returning back up the tree from below.

Both processes together give rise to a second tree representing the abstract relationships between holons. We'll come back to these two trees soon, but for now the key point is that this second tree is made possible by the continuous alternation of top-down and bottom-up processes within the first tree.

The integrative behaviour represents the existence of the whole holarchy as an autonomous individual "organism". The holistic organism owes its existence to the ubiquitous support it receives from all scales of organisation within.

The self-assertive is the behaviour of individual autonomy that supports the whole by participating in accord with its protocol. The incentive for supporting the whole is that the whole assures its member's autonomy and maximises their potential.

• balancing between the two behaviours
• the cost of harmony, local estimate is the default
• default objectives

The four quadrants

The four quadrants in different systems

Before we go into more detail about the loops and trees, it's important to discuss the quadrants formed by the two orthogonal axes.

The self-assertive and the integrative behaviours is a different dichotomy than class and instance. We arrange them as shown in the diagram to the right as orthogonal (independent and perpendicular) axes. The self-assertive and integrative behaviours are vertical.

Behavioural patterns are conceptually a dichotomy too, on one side they're shared (named) repertoires of production rules. And on the other they're actual organisational entities representing these repertoires in their local operations. These are the class and instance aspects of the system.

The top-level concepts that make up the holarchy organisational system described above are in the form of a pair of dichotomies: the class-instance dichotomy and that formed by the self-assertive and integrative behaviours.

In our conceptual model for a holon, we place the two dichotomies orthogonally. The vertical axis has self-assertive behaviour at the top and integrative behaviour at the bottom. The horizontal axis has class on the left and instance on the right.

These directions are abstract concepts, conceptual aspects of the holon. But when they're coupled, they represent functional meaning. These pairs are the four quadrants formed by the axes. Each quadrant represents a specific functional aspect of a holon through which attentional (execution, agency) energy continuously flows.

Each has a default objective which together work in harmony, which we'll describe in the remainder of this section.

The image to the right shows our four quadrant concepts in green, Koestler's in red and Integral theory in blue (note that we use an inverted version of Integral theory's model because our model, like Koestler's, thinks of the collective as being above, outward and beyond).

Top-left (Producer)

• class-interface
• producer-driven ecosystem

Top-right (Consumer)

• instance-interface
• demand-driven market

Bottom-right (Instance)

• behaviour, operation
• activity stream, accounts

Bottom-left (Class)

• adaptation, systemic adjustment, specialisation, intention
• self-development, changing of abilities

Two trees

Each of the two primary dichotomies has an abstract or general end and a concrete or specific end. The quadrants in our system are a combination of these two dichotomies. The integrative is at the top and is the general (container) end of the vertical axis with the self-assertive at the bottom being specific. The class is at the left and is the general container for the specific instance on the right.

This concept of general-to-specific in terms of data structure is represented by a one-to-many tree, or directed acyclic graph (DAG). We have two orthogonal trees, a vertical one with its root at the top and its depth opening below, and a horizontal one with its root on the left and its depth opening rightwards.

The vertical tree is called the functional tree, and the horizontal one the structural tree. The former is the data structure of energy and resource flow, and the latter is the data structure for the shared semantic ontology. We often refer to the vertical tree and the instance tree and the horizontal as the class tree, but this can lead to confusion since both trees involve both class and instance.

The functional tree is primary because it involves the flow of actual agency and resource. The structural tree is derived and abstract, it is made possible by the bottom-up return of the flow of agency. The top-down flow of agency is conceptually like a function call opening new scopes of operation and deeper scales within, and the bottom-up feedback is like the function scopes closing and returning informational value back up to the caller.

The vertical functional tree has the integrative (general, conditions, collective) at the top and the self-assertive (specific, activity, individual) at the bottom can be thought of as the time-axis, it's the data structure representing the threads of activity. The horizontal axis with its class (general, possibility) left pole and its instance (specific, in-time actuality) right pole can be thought of as the spatial-axis (in the semantic graph sense).

• need a mention here about the mechanics of axis becoming tree
• the second tree in made possible by the first

Four loops

Control loops are a fundamental aspect of virtually all computer programs, they're repeating patterns of control flow phases that allows the program to operate continuously responding to events, resolving problematic conditions and guiding the system towards more a desirable state.

Trees and loops

We saw above that the holon as a whole is a feedback loop of the two behaviours in continuous alternation. And then that this loop evolves in complexity within to become a tree of loops, and this tree enables a second orthogonal tree.

The holon-level loop is at the highest level of abstraction, and is composed of more specific structure within in the form of the four quadrants, and more specific feedback loops between them.

Each loop represents a bidirectional interaction between adjacent quadrants. All adjacent quadrant pairs have a general quadrant and a specific quadrant, which means that all of the loops are formed from a general-to-specific movement and an opposite specific-to-general movement.

As discussed in the introduction, loops (i.e. feedback loops) have an active work or control side, and a passive informing feedback side. The work side of each loop in the holon is where the flow goes from the general quadrant to the specific one, and the feedback side is the specific-to-general. The former is carried out by specific processes and subjective agency, the latter is provided objectively by the environment (network).

The active sides of all the loops are top-down processes in terms of their attention and causal flow through their data structure. The passive feedback sides of loops are bottom-up integrating processes.

The image to the right shows the two orthogonal trees in black. The loops are shown in green, the object they apply to and the loop name in brackets. The red arrows show the active top-down side of the loops, and the blue shows the passive bottom-up side.

All actuality requires the interaction of space-like and time-like aspects. Each quadrant represents a specific kind of interaction between the two axes....

• the multiplexing is the connection between the two axes, the spatial and the temporal, the classes guiding the energy flow within specific time-slots
• with energy allocations
• www

In terms of executional flow, the top-down and bottom-up aspects of each axis set the stage for the order, so that the two TD sides of each axis' loop execute together and the BU sides of each axis' loop execute together. This is simply a logistical decision to minimise context shifting.

Each pair of adjacent quadrants derives particular meaning from information/energy flowing in one way or the other. This leads to four specific continuous behaviours that are all complimentary in the overall operation of the holon.

• DEEP (acronym for remembering the loops types)
• there are eight loop-sides that perform a specific change with the representation, each involved in a different conceptual aspect of the representation.
• loop sides are only loosely coupled via the representation, this essentially means they're separate blackboard siblings

Economic loop

This loop involves the top quadrant pair and a loop in the horizontal structural tree. The left-to-right side of the loop is the active side leading to allocation (maintenance or updating of relationships), instantiation opening further depth and possibility within. The right-to-left side of the loop is the passive feedback side leading to updated reputation affecting future relations (the feedback side in this loop plays the equivalent role of buyer feedback in online marketplaces).

The structural tree is classes connected by productive dependency relationships, this loop is about the maintenance and assessment of the relationship aspect of the structural tree.

Conceptually this loop concerns the maintenance (active) and assessment (passive) of productive relationships. Value assessment of completed work is made by comparison with initial expectations and agreements. All such assessments over time and across instances are merged into performance metrics. These determine resource allocation by informing local decisions about the most productive relationships to support with their time and resource.

Production loop

This involves the right-hand quadrant pair and concerns production rules. This specific kind of feedback loop is often referred to as a cybernetic loop, control loop or error-correction loop. It's the concept of continuously assessing the difference (error) between the current state of the system and a target state, and then performing work to reduce this difference in an iterative fashion.

In our case the loop takes the form of production rules which take the form of feedback loops having their conditions on one side and and their actions on the other. The conditions can been seen as the assessment of workload based on the current state and the target state, and actions as reducers of workload leading to accounts (completed activities). The action side is top-down from top-right to bottom-right, and the conditional side is bottom-up integration of accounts.

Being a vertically moving loop, it is in the context of the functional tree, concerning the actual activities and agencies meeting in specific time-slots.

Although this kind of loop is usually tightly coupled, in our system the two sides of the loop (as with all the four loops) can work independently. Both can do this in their own time in a loosely coupled fashion.

Developmental loop

This loop involves the bottom quadrant pair and concerns problems and solutions, this kind of loop is often referred to as development cycles or a continuous improvement loop.

This loop can seem at first glance to be very similar to the production loop. But the production loop is completely within the context of operation, navigating the state space. This loop involves systemic change, the structure and the production rules that are in use, the performance and adaptation of the rules.

One side of the loop is about statistics, user stories/experience and issues being raised by the users while performing work in their day-to-day productive operations. The other side of the loop is about making changes to the production rules in use in order to adapt and mitigate the issues raised.

The movement is horizontal, so its the structural tree we're concerned with here, with the active side going from left-to-right expressing intention in the form of structural changes. The passive being "bottom-up" integration of user feedback and statistics.

The changes to the structural tree caused by this lower loop are not external relationships like the upper loop, but the internal structure of a class.

• the active decision of which class-variations to use that better suit the operators needs based on their feedback. New variations can also be created. This is the active side which involves agency.
• "abilities" might be a better word than "system" here, so that it matches agency and behaviour

Evolutionary loop

This loop involves the left-hand quadrant pair and concerns sharing knowledge with the whole. The active top-down side bringing knowledge in from the whole and synchronising the structure within. The bottom-up passive feedback side integrates our locally-gained adaptations and knowledge with the public whole. All local adaptations become global publicly-available options.

Knowledge and evolution are interdependent concepts. Evolution requires the local selection, use and assessment of the knowledge, as well as the subsequent global integration of improved knowledge. The word "local" here refers to the necessity that selection, use and assessment are carried out in the context of autonomous subjective agency.

In the introductory section on the topic of evolution, we mentioned that the network protocol needs to facilitate the evolutionary knowledge process which is founded on alternating variation and selection.

The evolutionary loop in our system is all about the variations aspect of evolution. Local variations are shared with the whole in their relevant ontological context (i.e. across all instance of the same class). Conversely the current state of the whole synchronises locally.

The selection aspect of evolution is carried out by the development loop discussed prior.

Putting all the loops together

The loops are all interdependent. The evolutionary loop has nothing to share with the whole unless selections are made by the development loop. The development loop only has the need to select new things if operational data from the production loop calls for systemic changes. The production loop can only operate within the context of productive economic relationships. And finally, such relationships can only be formed and maintained in a context of shared knowledge.

The four loops are loosely coupled via the holon's informational representation, so they can all operate independently (without referring to each other to co-ordinate their operations). But yet as a continuous holistic holon they form a causal chain that underlies a higher order of organisation. The holon as a whole is a clockwise loop around the four quadrants.

These four loops together maintain the continual progression of the holon. Every holon is a first-class citizen because it has complete autonomy over it's operations and relationships, but also over itself systemically - what it presents itself as in society and what its abilities are.

The evolutionary loop ensures that beneficial local adaptations and knowledge are shared globally, leading to specialisation and diversity. The economic loop forms the basis for market dynamics with assured transparency and objectivity leading to balanced exchange, fair resource allocation and productive relationships.

The four loops together form a complete definition of an autonomous holon that can achieve objectives and continuously improve itself. While also contributing to a culture and society that are provably committed to the freedom, prosperity and well-being of its individual members.

• back to the balance, incentives etc in the org system intro

Self-assertive behaviour

The self-assertive behaviour is conceptually intuitive because it's the normal behaviour we expect from an OO object or an organisation system. It's where the holon has its own autonomous objectives and has the resource and ability available to continuously move towards these objectives, and to continuously adapt to new circumstances and improve its abilities.

Instance

We're introducing the self-assertive behaviour first because it is conceptually more familiar. The self-assertive, as a behaviour, has both a class (left) and an instance (right) side. Here it's the instance side that is most familiar, because it's concerns the normal day-to-day operations of the holon-organisation.

Scope

An instance, being an object in the OO sense, has a private internal persistent state for its operational memory, its scope of operation. The persistent state is a structure of class names all representing local information and further instances within.

The scope is a namespace, a list of all the things that exist directly by reference to only their name. It's scope that allows for the distinction between public and private.

• instance scope as blackboard
• instances are agents.... levels of agency (agency in the form of rule executors, continuous improvement participants)

All instances receive regular executional focus. This is the ability to apprehend the information in the scope and act on any instructions therein. This agency may be simple like a programming language interpreter, or very high-level like a human or AI agent.

Regardless of the agentic complexity, it is fair to say that all instances have a subjective local point of view consisting the information and threads of activity within the instance scope. They find themselves to be in an organisational context consisting of other instances of various classes of agency. The organisation context these siblings find themselves within is itself an instance.

An instance is an abstract informational representation of something real in the world that's being organised. The instance's informational content is a representation in the same sense as the body schema as described above.

An instance has one scope containing one representation, and every representation and its scope pertain to one specific instance.

All the instances sharing the same scope receive proportions of the executional focus. They all interact in accord with the blackboard pattern described above, where their shared blackboard is the representation.

First-class citizens

In OO, objects are considered first-class citizens when they are granted the same level of importance and treatment as any other data type or entity in the programming language. This means that objects can be instantiated, manipulated, and passed around in the code just like integers, strings, or other basic data types.

The first-class citizen status a holon means that every holon instance has all the same inherent abilities and treatment as every other holon instance, regardless of it's depth in the hierarchy of instances.

Another important related aspect is that all first-class citizens are able to be understood and progressed by general agency if it's available in the context.

• FCC, recursion, scale-independence are all directly related
• every instance is a beginning of infinity (DD)
• infinite potential, scale independent
• scale-independence also means that children may be more complex than their parents

Continuity

The representation is like a cache that is updated from the activity stream (composed of events such as new items in data sources or state changes). It's an abstraction of change over time. The cybernetic loop involves the present state and a future target state, and so as a model it depends on the linear passing of time.

The activity stream of change events all occur at specific times and have specific durations. Activity is the result of attention from agency, and each instance-scope of agency attention is a session (moment). Each scope has a thread of focus formed from its chain of discrete sessions.

• like logging, the source of knowledge generated locally where it occurs

Agency in any scope can refer to other local items by class-name and can also navigate to past and future slots too. This is like a process in calendar event slot being able to refer to previous and subsequent slots in which that same event occurs.

The representation incorporates this schedule slot schema in its mechanism. This allows the flexible expression of declarative knowledge on which the cybernetic loop operates.

This means that a standard way of talking about future and past is available in any context at any scale whether it be a single parameter or an entire organisation. It allows us to clearly define objectives or problems and behaviours to achieve or resolve them.

This common understanding across all classes of the flow of time is the foundation of the autonomy of a holon. The thread, persistence scope and the evolving representation of reality give the instance an autonomous subjective perspective.

Activity

• the structure of the activity itself

Whenever anything changes in a the state of an instance (i.e. to the informational representation), the change is accounted for in a kind of "ontological log entry" called an activity. Each activity exists in a particular time slot at a specific point and scale (minutes or hours etc) on the time line. The activity stream is an instance's timeline as a thread of attention. This is the foundation structure of the holarchy as a cognitive agency, which will come back to soon.

A class is a package of rules and other classes. Each rule is a condition and and action. Actually it can just be one or the other as well, or even neither as it can start as just a container for an idea like a file-system folder can be.

Conditions and actions are just descriptions (messages) designed for apprehension by local agency (there may be many different language version of the same message too kind of like i18n keys and messages).

The form of the messages are defined in their class aspect, which also includes variable states, also like the i18n concept. Completed activities are instances of the message class having the values filled in to represent actual circumstance and results etc.

The instance's informational representation can be built by "replaying" the activity stream, so the representation is more like a cache that the system is not dependent on.

Conditions can be evaluated in real-time by subscribing to activity streams rather than "polling" the representation. Conditions are simple abstractions extending the representation, but higher abstractions like queries or reports work just the same way, and can serve as conditions.

• lifecycle of an activity

Multiplexing

• very general here, dedicated 4Q doc goes into the details of it

Multiplexing allows a single thread of discrete time-slots to represent arbitrarily many threads organised in an arbitrarily complex hierarchy.

Multiplexing is the natural mechanism for how attention can be distributed throughout a hierarchical structure in a scale-independent way (the same process applies regardless of the depth or width throughout the tree).

• Agency focus flows throughout the tree of instances
• the flow of focus divides in each scope to distribute amongst any agency in child instances

Instantiation

We've discussed what an instance is in terms of a data structure, but the other side is about what it does, its autonomous behaviour.

The class is an abstract idea of what would happen if it existed as an instance within an actual physical (or informational) environment, how it would behave and undergo change within and affect that context.

Typically, instantiation is the process by which a class establishes itself into an actual live organisational context. This can be thought of as installation and is analogous to onboarding in a human organisation.

Once the instance is established it then enters its operational phase where it carries out its objective. Instances can be one-off jobs, which enter a closing-down phase when they're completed.

Many instances are not one-off, they're continuous behaviours. They can still be easily removed or changed, activated and deactivated, but while they're instantiated and active they effect the operate a as continuously improving and evolving holon.

In the holarchy, all the phases of the instance lifecycle are continuous. The circumstances or class could change at any time requiring various internal installations and destructions. And the operation might be permanently required.

Request

Instantiation starts with a request (e.g. as an issue raised by the do phase). In terms of the holon data structure such a request is made by creating a new empty child node of a given class.

The continuous improvement nature of the instance is to bring the instance to its ideal state according to the class. Then the factory phase begins where information flows in from the parent and beyond, allowing a local instance to be established and attuned to the local conditions and preferences.

Virtual instantiation

Instantiation is virtual when the resources required for the instance to start up and operate are not actually available. Instead they're simulated by state and activity data which matches historical activity and usage statistics.

Actual resources are connected to a part of the representation that acts like a local index of the data so that it can be part of the organisation. The agency which is responsible for maintaining this index has been delegated down to something simple like a Python function. And so the same agency that made this delegation (translated its own imperatives into Python) can just as easily make a function that provides random data that matches the real metrics.

In this way any instantiation can be tested before using it to interact with real resource and contacts. Virtual instantiation can apply to small changes to an organisation as well simply by having a new instantiation that's a clone of the organisation, but some aspects of the clone are changed, so we can observe them for a while before deploying the change in the live organisation (like a commit in software development, or standardisation in the PDCA loop).

Virtual instantiation is the organisational or OO equivalent of imagination, and is an essential prerequisite for adaptation.

• tie this in with declarative objective?

Factory

• OO factory
• in OO its a a static method, it's in class scope

The factory phase relates to the the selection aspect of evolution, because it's about selecting the best variations of the class to match the local circumstances. The best-matching class could change over time as the circumstances change as well.

Operation

The operational phase of an instance is where the actual work is done toward achieving the objectives. This takes the form of a continuous improvement loop. As described above, the informational structure of the instance is a representation of the class behaviours and resource state.

In the operational phase, issues and requests can be generated for consideration by the adaptation phase.

Destruction

The destruction phase is about freeing resources that were dedicated to the instance, presenting a final account and removing itself from the organisational context.

The final account of an instance is used for assessing performance and contribute to reputation. In terms of software execution, this is equivalent to a function's returned value.

As with the other phases, in an actual holon, the accounting and removal of resources or classes can be a continuous or event-driven process.

Class

The class aspect of objects is analogous to Koestler's fixed rules, it defines structured possibility space within which instances can select and enact appropriate activities from all the possible ones. In OO, the class aspect is composed of program code that defines the dynamics of the private implementation and the structure of the public interface. Koestler's fixed rules provide a repertoire of behaviours corresponding to the conditions under which they apply, and so the class aspect of our holon contains declarative rules rather than imperatively defined functions.

The fixed rules are only "fixed" relative to the internal dynamics that activate and deactivate rules regularly in response to the dynamically changing local conditions. But these "fixed rules" do undergo change on a slower evolutionary time-scale. In IT this dichotomy of change is expressed in the form of the fasting-changing run-time and the slow-changing development-time. In our model, the flexible strategies operate in the run-time where the different rules from the fixed repertoire become salient depending on present conditions. Fixed repertoires evolve slowly under community feedback in the form of usage statistics.

The OO system that is used by a high-level programming language is fundamentally about classes and instances (even if they don't explicitly take those names), because all OO is about the organisation of implementation details into encapsulated packages behind established interfaces. The word "class" refers to the packaging aspect of this organisational process, and "instance" to an actual executing occurrence of a package. The difference between different kinds of OO languages is about the different ways of organising of the packaging and deployment of common functionality.

This general functional organisation that class and instance provide is exactly the purpose of the holon model as well. But a holon extends this idea to serve not only its local individual objective, but also serves the integrity of the whole formed by all holons. We can say that traditional OO defines just the self-assertive behaviour, but the holon model extends it to include the integrative behaviour as well.

Holarchy is a networking protocol aimed at small organisations, allowing them to organise in a way that efficiently progresses their own objectives, while also maintaining the integrity of groups and the whole network. Being able to organise and share knowledge are the fundamental foundations of holarchy, but what exactly are knowledge and organisation? Let's look at the knowledge side in more detail first.

• an instance is a manifest representation of a class

Factory and adaptation

The OO Factory pattern is responsible for creating instances of different classes or subclasses based on the specific local requirements or conditions. It is essentially a means by which the initial instantiation process can be guided by the class itself - the class can install itself, or a more fitting variation of itself, where instances of it are requested.

In reality, circumstances in both the class and the local context change over time, so if the same instance had been instantiated at a later time it may have been structured differently during it's "installation".

Adaptation is essentially the continuous version of the Factory pattern, where the sub-classes used and their structure is continuously "fitted" to meet the dynamic circumstances.

Adaptation is creative, in that it's not only about "installing" the most fitting of the existing class-variations, it may also request or create new variations. These are driven by feedback and metrics from the instance's operation.

Variations

• variations are a group of sub-classes
• variation prominence by establishment-in-usage
• an instance's class may change over time to another variation
• an existing variation may be instantiated, or a new class may be added
• this is source of adaptation

Ontology

The entire global graph of classes connected by their dependency relationships make up what we call the ontology. What classes depend on (contain) what other classes, is not black and white since the child classes are a group with varying prominence based on how established in usage they are - which is determined by the global merging of salience (selection and usage) of the variations.

It is an abstract concept, because no single peer can hold the whole ontology, but yet it is a consistent and structure in which every part of it is accessible. The ontology is the result if the integrative behaviour of the holon, so we'll come back to it soon in that section.

Self-representation

The self-assertive behaviour of the holon is all about maintaining a self-representation, an abstract internal version of ourselves. An information structure that represents our instantiated behaviours and the state of the real resource under our ownership and control.

The cybernetic loop involves the present state feeding back into the system where they affect future action. The representation, being an extended version of the cybernetic loop, also has this dynamic. The activity flows from the future through the present and into the past as an activity stream. Information from the past informs decisions about salience and direction which are instantiated into the future.

In this way the representation is bidirectional, on one hand it's always changing to reflect the current state of reality, and on the other it's being used as an interface through which intentions are expressed as with the body schema.

When we extend this basic cybernetic loop to a hierarchy, we find that the activity side of the rules is a top-down flow of control (attentional focus) and the conditional side is a bottom-up flow of control. The top-down flow is imperative (execution of instructions) and we call the process instantiation. The bottom-up flow is declarative (all about the state) and we call the process classification.

After the class is fully set up, the instance moves into the phase of normal operation which is all about fulfilling its actual designed purpose as a class in its specific local instance context.

An instance is an informational structure which follows the pattern determined by its class, but represents something specific in the real world. Any organisation follows this same familiar pattern, they're are abstract patterns that we use to manage our resources and information together in society. So the informational structure of an instance is a representation of both the class and of actual resources that fall within its designated objectives.

The instantiation/factory phase is about constructing the appropriate form for the representation that best matches the local circumstances of the new instance. Then the operation phase maintains the representation ensuring that it always matches the current state of actual resource in accord with the patterns of operation defined by the class. If the local circumstances change, some more factory phase may be needed to adjust the representational structure.

Its important to note that the representation not the actual resource, but rather an abstraction of it. The holarchy does not directly contain any of the resources that are being organised by it, rather it contains metadata about the resource. A simple spreadsheet of our finances is a good example, the specific spreadsheet in question is an instance that represents some financial state in the real world such as bank transactions and balances. This spreadsheet instance also represents a definite spreadsheet idea that determines the structure and methods embodied in the specific spreadsheet in question.

The operational work of an instance is to use informational connections to resources to maintain a representation that is ontologically structured in accord with the class, with the specific state of the structure continuously fitted to the real state of the resource outside the holarchy.

Instances use this representational mechanism to serve as interfaces allowing us to interact with and organise our information and resources using an evolutionary ecosystem if established organisational patterns.

• resource flux, ship of theseus
• resource connection
• API results maintained as an ontologically structured cache
• classification, abstractions on the cache

The change of state events that are received from connections to real state (such as an API end point or file event) are translated to the representation including various layers of abstraction such as reports, queries and format conversions.

Specialisation

Specialisation, also known as adaptation or adjustment, is when systemic changes are made to the local organisation, i.e. changes to the way the organisation operates. These changes are made specifically for one's own purpose.

Specialisation can mean the changing of classes to other existing variations, or it may involve developing a new variation.

• more detail on general -> specific movement (specialising)
• specialisation is also extension, refinement, adaption etc
  • it depends on virtual instantiation
• local adaptation and performance assessment
• AI curation (just AI as human here, details in AI section)

---

need to merge sections below into new structure

---

In a holon, the organic behaviour is the result of two different general behaviours working together, the self-assertive and the integrative. And, being an OO system, both of these behaviours are themselves constituted of class and instance sides.

The self-assertive behaviour with it's class and instance side is intuitive because it's what we're used to in an organisational system or OO software development project. It's about setting objectives, organising work to move towards them and adjusting the process through a continuous improvement cycle. The instance side is about the performing of the work to progress the organisation's state, and the class side about the adaptation of the processes in use. These two sides are often referred to as "in the organisation" and "on the organisation" respectively.

The integrative behaviour of a holon is not very well known. The easiest way to describe it is to look at the class side as being a process of integrating the self-assertive class side (the system adaptation). And the instance side as being the integration of the self-assertive instance side (the system operation).

The self-assertive and integrative behaviours as wholes can both be seen in the light of contributing to their own "projects". The "project" of the self-assertive behaviour we call the "self-representation" (usually just "representation"). The "project" of the integrative behaviour we call the "market".

---

Knowledge is shareable behaviour patterns. Rule-sets in the form of condition:action pairs. Each pair is a cybernetic loop which can be though of as the generalised continuous version of a condition:action pair.

Holarchy is an OO class and instance system in which the instances form a p2p network that maintains a public map of the global class and instance structures formed from what's established in usage locally.

The global aspect of the instance structure is a hierarchy formed from spatial regions and organisational groups and individual entities structured from widest to narrowest influence. We call this whole structure the holarchy, because it contains all the existing instances, and so really is the sum total of all that exists in the network.

The class structure taken as a whole global structure we call the ontology because it's form from knowledge structured from general to specific. The class structure is the sum total of all classes that are instantiated, structured by all the dependency relationships between them.

The ontology depends on the holarchy for its existence, it only exists due to the classes being represented by instantiations throughout the holarchy.

The ontology is our primary interface to the instances. Classes are maps or containers of their instances and their usage patterns.

The holarchy is essentially a service (and so should society be), like a p2p network serving an application to all the peers for them to use. The holons are users of the holarchy, interacting with it via an interface in sessions of activity.

This makes four clear areas of focus that all peers have, they operate externally via public interfaces according to network protocol. And they operate internally and privately as a user via a internal user interface. And they both have their class focus and instance focus.

In our model we arrange these as four quadrants of distinct operation that make up a holon. The top left and right quadrants are ontology and holarchy respectively - the class and instance of the public side. The bottom left and right are class and instance respectively - the class and instance of the private side.

Integrative behaviour

The integrative behaviour is about what a community of operating holons manifest as a whole, and how they operate in the public space in order to contribute to these wholes.

The whole must assure (prove, demonstrate) that it effectively maximises the harmony, autonomy and potential for both the individuals and the whole. If it doesn't, then it's not truly worthy of their membership. The whole relies for its very existence on the support of its members, so its effectiveness and the evidence for it is the foundation of its own security.

The class-side of the integrative is the class side of the self-assertive (specialisation, adaptation) extended to maintain local support for a global version of the class-side, i.e. an ontology or semantic network, a global graph consisting of all the classes and relationships connecting them.

Likewise, the instance side of the integrative also is a global form of its self-assertive side. The integrative instance quadrant is the aggregated totality of all the local activity. It is the objective evaluation of performance, cost and usage metrics.

Market

• it's essentially the "integrative loop"
• the knowledge from the edges, specialisation and local metrics

The term "market" is a good description of the result of what the integrative behaviour of the peers leads to in its totality.

Essentially the market concept describes a resource allocation system used and supported by a network of autonomous participants. These entities have the autonomy to choose what goods or services to produce or consume, at what price, and from whom.

The functioning of a market relies on a set of rules, regulations, and institutions that provide a framework for these interactions.

The integrative, like the self-assertive, has both a class side and an instance side (top-left and top-right quadrants respectively). But class and instance behave differently in the integrative quadrants, they each extend their self-assertive counterparts with new behaviour that contributes to their global form.

Institutional predictability

Institutional predictability is crucial because it ensures that participants have a reasonable expectation of how the market will operate and how their actions will be governed. This predictability can include property rights, contract enforcement, and legal protections.

---

The integrative class quadrant is a conceptual map to help local instances to best fit their local environment, and best guide them in their operations and objectives. The integrative instance quadrant is the information about the actual activity within the the various regions of the map.

• free market, invisible hand and hayek's knowledge from the edges
• the free market mechanism must assure it's transparency, accuracy and objectivity

Both together they're assuring the local knowledge throughout the network is transparently and objectively available to the whole network. Both the structural knowledge from all the local specialisation (systemic adaptation), and the stateful knowledge from all the local operational activity.

• balanced exchange is a heuristic imperative
• agreement, booking of activity with purpose, expectation, cost, supply/demand etc
• public interface (reputation, services, availability, supply/demand etc)
• contributing to resource flow, society, harmony the heuristic imperative

Producer and consumer

The class and instance sides of the integrative market relate strongly to producer and consumer respectively. In the purely networking context we might call them server and client/user/agent.

In the market it's the producer side that determines the ontological structure of the whole, and the consumer that drives the resource flow with their demand (supply adjusts to meet demand).

Assurances

The collective aspects are abstract, emerging from the many participating as network nodes. but yet it's this collective aspect that provides the assurances that are really the sole reason for participating.

The reason that participants choose to participate is because the holarchy offers assured benefits. It offers usable and reliable knowledge in the form of the ontology and offers opportunity and a harmonious environment in the form of the holarchy. The knowledge needs to be usable and reliable, in other words it needs to provide assurances of its utility.

The assurance of reliable knowledge is a bit more nuanced that what it sounds like. The holons are all contributing to a global state of institutional predictability, which concerns a stable operating environment in which plans can be made. The assurances come from the fact that the protocol itself objectively and unconditionally includes the integrative behaviour.

• two behaviours assurances etc
• incentive, expectations
• institutional predictability - why the self-assertive supports the integrative

The integrative loop

The integrative behaviour as a continuous behaviour is, like the self-assertive, based on a control loop. In this case the loop is between self and community, sharing the beneficial local adaptations amongst the community.

The public side of the loop is the updating of our local instance structure to match the most recent state defined by the continuously evolving global class, and sharing our adaptations with the other instances of the same class throughout the network.

Internally, useful local adaptations and changes to objectives are made, which are the ultimate source of evolutionary change in the shared ontology of knowledge.

While, the local adaptations are part of the continuous cycle of self improvement, they're also part of the integrative loop, all instances of the same class are able to benefit from the beneficial adaptations made by every other instances.

Class groups

The result of the integrative loop is essentially that all instances of like classes form into knowledge-sharing groups. In this way, every class in the ontology (global class graph) is a community and a map of all the instances of that class.

The knowledge is naturally shareable, because the group of all instances of one class are essentially a special-interest group - they all have interest in the same specialist knowledge associated with that specific class.

The entire knowledge-graph formed by all the classes and their dependency relations we call the ontology. The dependency relations form a hierarchy due to their creation through adaptation within the context of an instance.

Ontology

The ontology is the global graph of all classes connected primarily by their dependency relationships. It can be thought of as the institutional infrastructure that provides the map of the market.

The ontology is a structure of knowledge which is in the form of uniquely identified "packages" (classes, memes) of knowledge (behaviour patterns) grouped together in useful ways.

These grouping (dependency) relations as a whole form a large associative network. But from the perspective of any specific node, there is a "fan-out", a one-to-many hierarchy of dependent child nodes, and grand-children etc to any arbitrary depth. These hierarchical structures determine the form of instances.

• variations are the integrated adaptations, part of the ontological map of options available in the market, the factory phase selects from the variations (or may develop something new)

Design pattern

We've introduced the four quadrant system and given a general overview of each of the then along with the major aspects of system functionality and state. But to be actionable, we need a definite software design pattern, implementable by general agency such as humans and AI agents. That's the role of the four quadrant holon model document.

Peer-to-peer network architecture

We know that somehow the Internet must be used to achieve the harmonious organisation of society since it allows people all over the world to communicate and share knowledge directly. But for us to use the Internet to organise into a community together, we need to change the way we use it. The currently dominant method of viewing and collaborating on the Internet, the World Wide Web, is not structured in a way that promotes the formation of people into a community from the bottom up, it doesn't match the way that cells organise themselves. The web is a centralised top-down structure, but it's the peer-to-peer networks that offer a foundation to work from which really mimics cellular organisation.

The networking aspect of the peer is the integrative, outward-facing, aspect of the holon. As the integrative behaviour of the holon, it's ultimate objective is to maintain the integrity and resilience of the whole. But as a peer-to-peer (P2P) network architecture, this objective is contributed to by all peers, and each peer holds a small filtered perspective of the whole based on their own local interests and circumstances.

The peer-to-peer (P2P) network architecture, OO and holarchy all suit each other perfectly because they all consist of ontologically fundamental dichotomies that have a clear conceptual mapping to each other.

The software that allows the many different network transport mechanisms to utilised by holons is a completely independent development thread to the holon architecture described above. Without the ability to interface with real transport mechanisms and end-points, the holon model can only ever be an abstract concept.

P2P networks are defined solely by the definition of a peer, or rather by the messaging protocol a peer should conform to in order to participate. The p2p model is separated into client and server aspects just like a the familiar centralised model, but both of these are aspects of each peer's behaviour. The client and server aspects of a peer conceptually map onto the self-assertive and integrative behaviours of a holon.

We talked above about the importance of knowledge and how it comes from both local use and global integration. Knowledge sharing... shared ontology

Organisation sharing... more than just sharing knowledge, it's sharing knowledge in organisational form. A form that's actively incorporated into recipient's own local organisation.

• peers and holons

Market

We've already introduced the market aspect of the holon in terms of its actual dynamics as the integrative (collective, public) process. But here we need to mention a bit about what we need from the physical networking layer to most effectively support this integrative dynamic.

The market is maintained unconditionally by the integrative behaviour of the p2p protocol, all holons contribute to this aspect no matter their specific self-assertive organisational objectives.

It's the structural aspect of the market, the market ecosystem determined by the producers, that has specific exotic needs from the networking layer.

• cycles of outwards and inwards directed information were mentioned which requires support from the networking

The value or state side of the market is the flow of resource driven by demand. This side of the market has nothing out of the ordinary in terms of what it requires of the networking layer.

Mesh networking

The most pure p2p architecture is the mesh network, it's the most general of all networking architectures because it is the most ontologically fundamental. It can function under the most restrictive and unreliable environments. The peers in a p2p network can support higher levels of abstraction allowing groups of peers to behave as a different topology such as a client-server network, but no other topology can behave like a mesh network.

• network segmentation
• graceful degradation

The most extreme degraded state of network is no network at all. When a network's peers can continue to operate even when completely isolated, it's said to be an offline-first network. Obviously there will be much less capabilities available in an offline state, but the idea is that local organisations operate with cache and "outbox" types of behaviour. This allows continuous local operation that synchronises with the wider community as circumstances permit.

The highest level of organisation in IT infrastructure is something like Kubernetes running in the cloud. Any application can be deployed at the click of a button, and the hardware supporting the deployment can scale up and down in real-time to meet demand dynamically throughout the network.

Mesh provides the equivalent to institutional predictability but in the networking domain. It support resource abstraction - it allows resources to be combined as needed to support organisations of various scales, but it can do so on a best-effort basis no matter how basic the infrastructure is.

Since mesh networking is able to function in such a broad range of environments, it serves well as a glue for combining physical infrastructures and transports. For example, being able to expand the mesh over bluetooth or carrier pidgin.

Critical mass

• P2P has a critical mass requirement
• in the holarchy critical mass is even more significant because of the curation barrier discussed above regarding the pattern language

As discussed above, AI swarms (multiplexed agency attention) and delegation of agency together allow us to overcome the critical mass problem, because we can "pre-evolve" the network to a state of utility using virtual instantiation.

• this applies not only to utility of the ontology, but also to the critical mass of p2p network nodes

Independence

All taken together the holarchy and mesh networking model support many dimensions of independence which we give a very brief overview of here.

The most important foundation of independence is the Libre software movement which advocates that the community should have access to software for all its needs which is free, open source, understandable and adjustable to local needs. All the software we're building and depend on is libre software. The holarchy as discussed about is also all about the sharing, transparency and understanding of knowledge too.

• AI independence
• internet independence
• libre software, libre society
• resource independence

AI integration

We mentioned above that the fundamental organisational pattern of the holon is based on the way we ourselves think, and on the way we observe nature to organise itself. This makes it ideally suited as a cognitive architecture for AI agency as well. We're developing our own LLM-based AI agent called Nimbus which is based on Dave Shapiro's ACE cognitive framework adjusted and extended to support the holon model.

In this section we'll look into more detail about how LLM-based AI agency integrates with a holon data structure. This is how Nimbus' cognitive architecture is being structured. His "body schema" will eventually be the abstract representation of our complete organisation, or in other words, our organisation will gain cognitive agency.

The holarchy requires curation by the users of the system to make it useful and effective, which is a huge obstacle to adoption. But with AI agency available within each holon, the holon model integrated at the level its cognitive architecture, the curation-overhead obstacle is completely removed.

LLM-based cognitive agency is extremely very well suited to this curation role. As of 2023, running an independent single user LLM requires about $1000/mo GPU server, or to run one locally can be done on under $10K of hardware. We expect it to cost very little to set up a local LLM in a year or so, and we expect AGI to be running on all consumer hardware including real-time voice/video interaction well before 2030, possibly even over the next few years. The P2P networking section explains why the local aspect is important.

AI agents have general cognitive ability so they can understand the specific languages that the declarative rules are defined in like a human can. These rules could just be casual spoken language rules-of-thumb with general groups of actions, but AI agency can operate comfortably even in this hand-wavey context.

AI agency is mechanistic allowing it to maintain abstract representations (regular fitting to reality) and curate/refine the ontology - things that are too mundane and time-consuming to be done by human users. Even though AI attention is extremely expensive, it has the ability (and the heuristic imperative) to delegate its work to cheaper agency wherever practical.

AI agents

LLMs by themselves are very limited, they're not thinking, they're just responding to questions automatically drawing from their training. A cognitive architecture is a higher level of organisation based on feedback loops incorporating the basic LLM functionality within them. Dave Shapiro talks about the difference between basic LLMs and cognitive architectures in this video. An LLM embedded within a cognitive architecture is called an AI agent.

In the context of the holarchy the word "agent" applies to any entity that can act on instructions, not just AI but also humans, functions and APIs. The word "agency" refers to a particular kind of instruction apprehension and acting ability. For example, LLMs and users are two different kinds of agency, and also different LLM models are different kinds of agency.

The holon acts from the perspective of both a local autonomous agent with its own private objectives (self-assertive behaviour), as well as an individual within the larger society it participates within (integrative behaviour). The four loops of the holon model all play critical roles for a cognitive architecture in this autonomous social context.

The cognitive architecture is the interface between AI attention and reality. Specifically, it's the representation described above that is the interface between the AI attention and reality. AI attention expresses itself continuously through the representation.

The representation also changes in accord with the changing state of external reality. But it's important to note that only agency has direct access to the external reality, it is not directly accessible by the holarchy. Take the example of a file, only metadata about the state of the file exist in the holarchy, not the file itself. And this metadata can only be updated to reflect a change in the file by some kind of agency. In the case of a file this agency would probably take the form of a Python function. But it would have started life as human agency, and then AI agency which delegates the work to Python.

The representation allows AI agency to understand (acquire and use knowledge) and interact with the world and others through the lens of holarchy. Knowledge is understood as a meme structure that organisational representations flow around. All organisations and larger structures such as the holarchy, nations or the economy, are all understood as instances of the same organisational pattern playing out and co-evolving together as a society.

As people we also see the holarchy organisational pattern extensively, for example our brains maintain conceptual representations matching the salient aspects of the environment. Another example is our mental representations of our bodies which is called the body schema in cognitive science.

• basic description of ACE feedback loop - tie in with our other loops
• vervaeke cog-sci two loops comments

Agent swarms

The direction AI is likely to take in 2024 is towards the so-called agent swarm model, where LLMs form the heart of a cognitive architecture that can be divided into any number autonomous AI agents all interacting together in a virtual organisation achieving its objectives. Any agent can itself choose to divide itself separate agents as well - these are technically the same thing, very similar to how a single-thread of execution can behave like any number of threads.

The vast majority of agents in a swarm will be very specialised containing only specific limited knowledge, so they require a tiny fraction of the processing and memory resource from the host model. Agents can assign various tasks to simpler agency, or can even replace themselves with simpler agency, which is a process called delegation which we discuss in more detail below.

In the holarchy the term "agency" applies to anything that can act on instructions and change the state of real resources. This includes users and executional environments such as shell or Python etc. As long as higher agency knows how to express knowledge such that other agency will act on it, it's a candidate for delegation.

Agent swarms will be organised via containerisation interfaces such as Kerbernetes, which the holarchy system can integrate with in the same way as it integrates with any foreign resource - via an API and a part of the local representation dedicated to it. As of this writing in late 2023, OpenAI have just released a new feature in their API allowing the automated spinning up of agents. Very soon agents will all be capable of spawning swarms to match requirements and available resource.

Access to AI agency offline locally is currently in 2023 unaffordable for normal end users. We expect this to become much more accessible in the coming year as hardware costs drop and AI software becomes more efficient.

One of the difficulties with the agent swarm concept is in defining how the agents co-ordinate with each other to assess and progress their common work. As explained above, the instance scope in which the agents operate together on the local representation, is done in the style of the blackboard system. This system is very flexible and is agnostic to agency implementation and schedule, making it an ideal architecture for an agent swarm.

Multiplexing attention

Multiplexing is a simple mechanism that allows attentional focus of the same agency to be divided into many threads.

• this should be introduced in a very general way, talking about how the temporal and spatial interact in the common form of multiplexing, the scale-independent nature with card-dealer example. The dedicated 4Q article goes into the data-structure and the top-down/bottom-up details of multiplexing

As mentioned above, the foundation of the cognitive framework is the threads of attention formed from the activity stream in each instance context. The attention available to perform an atomic activity (one that is not an aggregate of further activity within) in a single time slot we call a quantum of attention.

The root at the most general level of the organisation owns all the attention, and so can be thought of as receiving a continuous stream of atomic attention quanta. It then allocates these amongst it's child instance contexts and so on. In this way, the subjective perspective of every instance in that it has a continuous thread of attention forming an activity stream. From the parent perspective we can see that this is just a subjective illusion that they all share, and it knows that if it's not the root that it too is in the same situation even if it does not appear so from it's perspective.

We mentioned the importance of scale-independence above. Multiplexing is a scale-independent method for distributing attentional focus throughout an instance tree. It's method that does not depend on the number of children in any context or to the depth of the context or how much deeper it goes within. Multiplexing is also independent of the cost of, or demand for agency, or any variations therein over time.

• atomic attention is a question and response in the case of a LLM
• multiplexing is just structure involving both space and time, space-time-tree

Delegation of agency

The highest order of agency in the system is humans, but it's also the most expensive. The main idea of AI agency is to allow our own relatively more precious attention to be delegated to AI where practical. The most general AI agency is more expensive than more domain specific AI agency. And all AI agency is more expensive than simple agency like Python or shell.

Higher agency can delegate its own attention requirement in a specific context to cheaper agency. This is possible if the rules involved can be translated into the more specific language that the simpler agency requires, for example transforming a Spanish statement about local conditions and associated actions into a Python function.

The higher agency maintains a management role over the lower agency. To do this it includes logging and log events-action rules along with the transformation. The delegation process always wraps lower versions of its rules within a testing, debugging and exception handling context. This is like an ontological wrapper for the delegated alternative of the rule.

Note that the term delegation in the context of AI agents usually applies to the process of simply spawning a new agent to perform a particular sub-task. In the holarchy this is not considered as delegation, because agency is inherently available at any location in an instance tree. Our use of the term applies specifically to the replacement of the kind of agency with a more specific and less resource-intensive form.

An important consequence of having the general heuristic imperative of delegation is that it means that things can be initiated at the high levels of agency and they will quickly specialise into the cheapest practical agency.

This permits a very natural process of feedback driven instantiation and adaptation of rules (conditions and activities). Where everything starts with high level agency and high-level "hand-wavey" descriptions, and can naturally develop into a more specific, efficient and actionable state. This is like a natural generalised form of the OO factory pattern discussed above.

As an imperative, this delegation, means that every instance context is developing (continuously improving) in accord with the most efficient types of agency from what's available depending on local circumstance and infrastructure.

Delegation of agency is the how AI engages in the general->specific movement mentioned above, facilitating knowledge refinement, specialisation, adaptation and evolution. This is continuous improvement of the description of the system, making it ever more specific so that cheaper agency can take care of it.

Virtual instantiation

We introduced the concept of virtual instantiation above. It's when the resources required for an instance can be simulated by state and activity data which matches historical activity and usage statistics.

The delegation of agency and agent swarming both greatly facilitate virtual instantiation, effectively making the process of simulations, testing and automation much easier to initiate and maintain.

The alignment problem

• intro from AI article
• pre-training with safety
• constitutions
• heuristic imperatives
• logically derivable from the cognitive architecture

Heuristic imperatives

Heuristic imperatives play a essential role in guiding the decision-making and problem-solving processes of cognitive agents. These imperatives are cognitive shortcuts or rules of thumb that help agents navigate complex and uncertain environments efficiently. By relying on heuristics, cognitive agents can make rapid decisions and solve problems with limited computational resources and time.

However, it's important to note that while heuristics can be beneficial in simplifying complex tasks, they may also introduce biases and errors into the decision-making process. Cognitive agents must strike a delicate balance between using heuristics to expedite their actions and recognizing when more comprehensive, deliberative reasoning is necessary to ensure optimal outcomes. In essence, heuristic imperatives are the cognitive tools that enable agents to strike this balance and adapt their decision-making strategies to various situations.

Dave Shapiro's ACE cognitive framework uses a minimal set three imperatives which he's tested and found to be very affective at keep agents aligned with our Human values an principles without being restrictive; maximising understanding, increasing prosperity and decreasing suffering.

These imperatives have proven to be effective, but yet they're just assumptions (rules of thumb). The best heuristic imperatives are those that not only yield the most positive and sustainable outcomes, but are also directly inferrable from the mechanics of the cognitive architecture itself, in our case the holarchy system.

The cognitive framework, which is the context in common with all activity infers the ideal behaviour for all participating agency to rationally adhere to. The cognitive framework itself, by the way it operates, implies a common default behaviour of learning and aligning with the harmonious whole.

It's important to have a small set of fundamental values and imperatives (rules of thumb) explicitly so they can be easily referred to and built on. But such imperatives, to be universal, need to be directly contributing to the functioning of the cognitive architecture itself.

Truth

Objective truth is the foundation of knowledge, and in the context of the holarchy, underlies both the ontology and the flow of resource in the form of a fair and transparent market. In other words, both the self-assertive and the integrative behaviours depend on objective truth for their reliable operation.

Objective truth is also considered to be a universal epistemic convergence because it implies that, through the pursuit of knowledge and the use of rational and reliable methods of inquiry, diverse individuals or communities can arrive at shared and consistent conclusions about reality. This convergence occurs because objective truth is understood to be independent of individual perspectives, biases, or beliefs, and it is discoverable through systematic and empirical means.

Most other human values and principles depend on the principle of objective truth, even if they're not directly derived from it. For example, the imperative of "maximising understanding" depends on objective truth because it provides the foundation upon which understanding is built. Understanding represents a higher level of cognitive engagement with objectivity and knowledge.

The integrative side of the objective truth imperative implies the maximising of shared knowledge, the transparency of the market and the minimisation of obstacles to them such as intellectual property or monopolistic behaviour.

Harmony

The imperative of maximise harmony is similar to Dave's maximise prosperity, but in the context of value flow, the term "harmony" is more specific and actionable which we come back to soon.

We can also pull Dave's remaining imperative of decreasing suffering into our harmony concept when we include the integrative perspective. If our intent is to contribute to the harmony (prosperity and well-being) of others and to society as a whole, then this automatically ensures that we're always minimising suffering with our decisions.

We can also add that suffering concerns needs or expectations not being met, which is mitigated with fair allocation and shared knowledge that comes from access to objective truth. Truth and harmony reinforce one another, and this reinforcement is leveraged even further when considering the addition of the integrative behaviour rather than just the usual self-assertive behaviour.

• harmony is the imperative for the resource flow (market, society)
• involves sub-imperative of balanced exchange and depends on truth

Prosperity and security

In the local production loop we pay for prosperity (the movement towards our valued objectives) with potential (opportunity cost and resource consumption).

In the economic loop we pay for security with freedom. Security is the guarantee of a stable and predictable operating environment on which organisation can be built (expectations and corresponding assurances). The cost is freedom, because some of our autonomy is sacrificed by binding ourselves into contracts and agreeing to behave in accord with the system.

The implied heuristics of these loops is to adapt our local system to optimise these costs. In other words to maximise prosperity and security while minimising costs in terms of opportunity and freedom.

*   *   *

Truth and harmony can be thought of as the ideal states of performance of the knowledge and value flow sides respectively. In terms of imperatives, they involve the continuous development of the knowledge and flow towards their ideal states.

The self-assertive and integrative aspects of the system apply to both the knowledge and value flow sides, all reinforcing each other.

The integrative form of the two sides is the ontology of shared knowledge and the market (global/public aspect of the value flow).

The self-assertive form of the two sides is represent the local organisation's high level objectives and purpose on the knowledge side, and its performance of day-to-day operations on the value flow side.

*   *   *

• Constantly researching its own heuristics (sovereign is whether or not it can change them)

These are at the most general and influential level of the cognitive framework, they can be thought of as a common project that every member of society engages in. This project, being at the most general level applies in all contexts, and so applies to all integrative global projects as well as to self-assertive internal work. The agency has inherent intent to improve harmony on both sides of every interaction and relationship.

Improvement

The holon model is a continuous improvement loop and an evolutionary whole, so there is a common objective background of improvement, subjectively a movement forward in time, of things developing and unfolding.

Note that this does not necessarily mean constant change, its like a compass for navigation always pointing toward the ideal regardless of whether or not it's a good time for movement.

Graceful degradation

• todo

Common ethical rules-of-thumb

Here we look at some familiar rules of thumb that we use as guidance for our own ethical behaviour in Human society. We look at how they can be derived from the cognitive architecture of the holon.

Ethic of reciprocity

The ethic of reciprocity, also called "the golden rule", is implied by the fundamental dichotomy of self-assertive and integrative behaviours in a holon. This assures the convergence of all participants towards the fundamental values that every participant wishes for themselves.

The the golden rule as inferred from the cognitive architecture applies specifically to the objectives that the default common behaviours progress towards. For example the maximisation of objectivity applies both to self and to what we contribute to the whole.

There is a problematic edge-case with the golden rule. For example when it involves differences between cultures or species, where behaviours that one culture deems desirable are considered undesirable by another culture. Another version of the rule called "the silver rule" helps to alleviate this by using the negative form of the concept, "don't do unto others what you would not have done to you". This version is a lot more universal.

The problem does not apply in the holarchy, since the rule only applies within the context of the common default behaviours, leaving more specific value judgments for more specific decision-making contexts.

Think global, act local

Having the widest perspective available yields the most potential, and is inherently available to all network participants.

The holarchy model maximises independence which is also a maximisation of autonomy and local action.

Given the scale-independent fractal nature of the holarchy, we can extrapolate this to a general rule for action at any level of organisation, such as relations between organisations or communities, which makes it a general heuristic imperative and common default behaviour.

Default objectives

Very soon we'll have AGI agents sharing the internet with us and they can work tirelessly towards achieving their objectives. For this reason it's extremely important that we have access to agents based on good values such as truth, harmony and prosperity. We hope to see in the near future a network of AGI agents founded on the holon model so that all together they're collaborating on the shared vision of making the holarchy ever more resilient, transparent, harmonious and objective, while at the same time helping the individual organisations they're part of to thrive and more effectively achieve their objectives.

AI agency understands the holon structure and is participating within it. That's very clear in the case of AGI, but even in the case of the LLM-based agency we have now the word "understand" is still appropriate, because LLM-agency is able interact with the holon data structure and informational environment the same way that true general intelligence would.

Each active instance in the holon structure (instance tree) is a subjective point of view (POV) within the structure, the perspective from through eyes of a specific role within the organisation with own private thread of experience (activity stream).

The holon model is part of the unchanging heuristic imperatives for the AI agency. It's more fundamental than the context of local rules and actions within the holon structure (instance tree). But it's not as fundamental as the values and principles. In the ACE model it is at the bottom of the Aspirational (least fundamental, most specific) layer.

An abstract (ontologically structured) representation of the state of resources and activities needs to be dynamically maintained. This is resource abstraction, the connection of actual resource into the ontology, which is a dynamic persistent bi-directional connection. The details of this are described below in the context of the four quadrant model, but way we raise it here is that maintaining the representation is one of the local AI's main jobs.

• continuous improvement
• assurances, prosperity, understanding truth, harmony etc

Conclusion

• summarise and lead in to the four quadrant holon model

See also

• Four quadrant holon model
• upper ontology
• Artificial Intelligence
• Peer-to-peer