Stafford Beer Revisited

Toward Digital Twins (David Carr)

Nowadays Stafford Beer’s foresight of the Viable system model (VSM) can be realized through digital twins and ontologies.

Viable Systems & Digital Twins

Stafford Beer’s Viable system model (VSM)

Stafford Beer is best known for bringing cybernetics to management, both theoretically, with his Viable system model (VSM), and practically, by trying to implement it with computers and to apply it to Chile’s economic planning under Salvador Allende. Despite the foresight of the project, Beer and his team could not overcome the technological impediments of the time, not to mention Chile’s military backlash. But Beer’s vision is now taking a new relevance for enterprises immersed in digital environments.

VSM & Digital Twins

For enterprises competing in digital environments change is the rule of the game and their architectures must undertake emerging as well as planned evolutions driven respectively by:

  • Unforeseen and/or forced changes, possibly at odds with the state of their organization or systems
  • Designed changes of their organization or systems, possibly misguided or out of sync with changes in environments

To that end enterprises can use ontological prisms to turn their architectures into digital twins ensuring a full transparency and interoperability of data (extensional), business (intensional), and system (symbolic) models.

Viable systems & Digital twins

Such digital twins can be seen as a comprehensive and actionable realisation of Beer’s Viable system.

Viable Systems & Cybernetics

Change being a matter of time, Viable systems must deal with a shifting balance between extensional assessments (which improve with time until becoming redundant) and intensional commitments (which bear the costs of missed opportunities if kept waiting too long).

Digital Twins & Cybernetics

The challenge for Viable systems can thus be best summarized through the cybernetic concepts of osmosis, homeostasis, and entropy.

Osmosis

Osmosis is the exchange of operational data between systems and their environment; given the inherent porosity of digital boundaries, data mining can provide comprehensive and timely accounts of external events, and process mining enables fine-grained tuning of operations.

Osmosis between digital twins and environments

One step further, enterprises can bolster lean and just-in-time engineering processes through the direct integration of (digital) business logic into software (e.g using rules engines), thus reinforcing shortcuts between observations and decisions whenever intermediate analysis is not necessary. The integration of engineering and business processes will enhance interoperability between platform specific (PSMs) and operational models, and consequently enterprises’ homeostasis.

Homeostasis

Homeostasis is the ability of organizations and systems to align their governance to changes in their business or digital environments.

These objectives can be met directly through a specific mapping of applications problem to system solution spaces (counter-clockwise path), or indirectly through business processes and domains spaces to system solution spaces (clockwise path). When the alignment of problem and solution spaces is specific (typically standalone applications), mappings can be directly achieved through taxonomies and database schemas; but when problem and solution spaces are set across (typically through shared business domains and functions) ontologies are necessary for the intermediate mappings of meanings (communication) and domains (organisation).

Balancing emerging and planned changes

In theory Viable systems could take advantage of the duality and actionability of digital twins to iron out discrepancies between business and systems realms, e.g.:

  • Planned changes, comprehensive and seamless model-based engineering applied to shared business functions as well as specific applications
  • Emerging changes, direct and fine-grained embedding of database schemas and business rules in applications

Yet, cybernetics point to the intrinsic loss of information in exchanges between systems and their environment, inducing a misalignment between symbolic representations (maps) and actual environments (territories).

Entropy & Complexity

As understood in cybernetics, the entropy of a system is the quantum of energy that cannot be converted into mechanical work. Formalised by Shannon in terms of leakage along communication channels, entropy has become a pillar of information theory.

Reset in economic terms, entropy can be understood as the quantum of data that cannot be explained by information models and/or put to use as knowledge. It ensues that for enterprises in competitive environments, changes in entropy stem from communication, namely how exchanges affect the relevancy of models; when enterprise architectures are embodied in digital twins the issue can be dealt with through ontologies.

Ontologies are often reduced to conceptual models enclosing stable and unambiguous meanings; but that understanding goes against what happens to meanings in shifting, competitive, open-ended, and overlapping business environments. From a cybernetics perspective, enforcing a comprehensive and continuous semantic alignment for terms used in data, business, and system models, would both reduce ambiguity and variety, and consequently the versatility and plasticity of systems; hence the benefit of ontological levels of indirection:

  • Taxonomies, for the alignment of variants (facts) with database schemas (categories)
  • Thesauruses, for the alignment of meanings between facts (data and applications) and concepts (business processes)
  • Ontologies, for the alignment of business (conceptual) and system (categories) domains

Entropy thus appears as the counterpart of extensional and intensional complexity, the former set by environment and operations, the latter by business and organisation.

Entropy as misunderstanding

In principle Viable systems should achieve a balancing act reaching an optimum somewhere between:

  • Zero entropy, maximum complexity: each micro-state in problem space is mapped to a macro-state in solution space
  • Maximum entropy, zero complexity: a single macro-state in solution space for the whole of micro-states in problem space

In practice, given that there is no effective scales for complexity or entropy, appraisals must be carried out for changes in ambiguity and variety; to that end Viable systems could rely on ontological prisms and knowledge-based decision-making processes.

Viable Systems Agency

Taking into account the limitations of the technologies available at the time Stafford Beer’s vision of VSM capabilities and agency has retained some relevance, especially when considered in terms of cybernetics.

Subsystems & Capabilities

Beer’s original outline of subsystems is meant to mirror human brain and nervous system capabilities:

  • Autonomic nervous system, for the control of primary (non-symbolic) activities
  • Cognition, for symbolic processing and reasoning capabilities
  • Consciousness, for prospective thinking, judgement, and decision-making

Taking advantage of ontological prisms four basic cognitive capabilities can be introduced for observations (facts), reasoning (categories), judgement (concepts), and experience (facts).

These augmented capabilities can then be aligned with the kind of ambiguity attached to “Algedonic” events:

  • Unambiguous events can be processed by primary capabilities (a)
  • Extensional ambiguity stems from discrepancies between events and known representations; its processing requires reasoning but not judgment (b)
  • Intensional ambiguity stems from misalignment of objectives and events representations; its processing requires reasoning as well as judgment (c)
No ambiguity (a), extensional ambiguity (b), intensional ambiguity (c)

That framework can serve as a reference for control and decision-making policies.

Control & Entropy

For Viable systems controlling entropy entails real-time assessment of Algedonic events; on that account John Boyd’s Observation-Orientation-Decision-Action (OODA) model appears to be Viable systems’ perfect complement. Rooted in Boyd’s experience as fighter pilot, the OODA relevance comes from a seamless integration of data capture and elicitation with command and control processes, an integration now boosted by AI and ML technologies; Viable systems agency can thus benefit from a pairing of OODA steps with ontological prisms:

  • Observation: separate the chaff (data) from the wheat (information)
  • Orientation: identify causations, risks, and opportunities
  • Decision: weight risks, make commitments, plan operations
  • Action: carry out decisions, make adjustments, collect experience
Pairing OODA with Ontological prisms

Events can then be monitored according to the ambiguity of observations (data), orientation (information), or decision (knowledge):

  • No ambiguity: events can be directly interpreted and feed orientation
  • Extensional ambiguity: uncertainty about events can be reduced by additional observations until further delays increase complexity and thus impair orientation
  • Intensional ambiguity: uncertainty is rooted in orientations and cannot be reduced by further observations

The viability of systems thus depends on their ability to manage ambiguities, and plan accordingly .

Learning

Viable systems must survive in changing environments, and to that end they must learn from experience.

Practical vs reasoned learning

That can be achieved directly from practice, using judgment to form beliefs from observations (clockwise), or indirectly by using reason to frame observations and forming beliefs from there (counter clockwise).

Survival will then depend on applying practical and reasoned learning to algedonic events, and to plan accordingly.

Planning

Planning is at the core of systems’ survival. Given Viable systems learning capabilities, their planning should be defined by what can be known about algedonic events:

  • Deterministic planning: when unambiguous algedonic events (like suns’s and moon’s) can be directly ascribed to relevant information and put to use as knowledge; such planning can be frozen as procedures (a).
  • Stochastic planning: when algedonic events involve random variables whose potential range can be accounted for by existing symbolic representations; such planning can persist as long as the reliability of observations can be improved or until the “last responsible moment,” when further delay would be detrimental (b).
  • Strategic planning: when algedonic events may involve open-ended developments which may or may not be accounted for by existing symbolic representations; such planning can persist as long as risk-management schemes can cover for ill-fated turns of events (c).
Viable systems planning should be framed by horizons: deterministic, stochastic, or strategic

Borrowing from Donald Rumsfeld’s often quoted taxonomy, planning can thus be defined by things we know (deterministic), things we don’t know (stochastic), and things we don’t know we don’t know (strategic).

Further Reading

Kaleidoscope Series

Other Caminao references