Systems View

How does anything stay on course when conditions keep changing and no one can see everything? Reasoning has become that kind of system.

What is now usually called a systems view grew out of cybernetics and general systems theory: the study of how systems observe themselves, regulate behaviour, and adapt through feedback. It is not just an analogy here. The word “navigational,” in this research programme's own Navigational Coherence, comes directly from cybernetics' own root, the Greek kybernetes, a steersman. See Hand on the Rudder for the fuller story of that connection.

Systems That Learn From Their Own Behavior

Cybernetics emerged in the mid-20th century as a field studying communication, control, and feedback in complex systems. Norbert Wiener, and later the second-order cybernetic thinkers, examined how systems maintain stability by observing and adjusting their own state.

In a cybernetic system, feedback lets the system compare its current behaviour with desired outcomes and correct the difference. Without observable state and feedback, an adaptive system cannot regulate itself. The insight applies to machines, and equally to organisations, institutions, and knowledge systems.

The Observability Problem

For a system to correct itself, it must be able to observe its own state.

Many reasoning environments today produce large volumes of output — documents, reports, analyses — while the reasoning that produced them remains difficult to observe. When reasoning trajectories disappear across conversations, tools, and collaborators, the system loses the ability to evaluate and refine its own reasoning.

Cybernetics has a name for this: a loss of observability. Without visible system state, feedback becomes unreliable and corrective learning becomes difficult.

Requisite Variety, and the Model That Must Stay True

W. Ross Ashby, alongside Wiener one of cybernetics' founding figures, gave the regulation problem a precise form: a regulator can only handle as much variety in its environment as it has variety in its own possible responses. A fixed, one-size-fits-all response cannot regulate a genuinely varied situation. This is Requisite Variety, worked out in full in Architecture's Systems Perspective.

Conant and Ashby's good regulator theorem sharpens it: a system that regulates well must contain a model of what it regulates. Read for reasoning, the model has to stay true to the system as conditions change. That is the cybernetic form of what the Problem page calls calibration.

Applied to AI-accelerated reasoning: a governance system that can respond in only one way cannot regulate the actual variety of situations that reasoning produces, and one whose model of the situation has drifted is regulating something that is no longer there. This research programme treats reasoning-handling capacity as something that has to match the variety it actually meets, not the variety anyone assumed in advance.

Reasoning Artifacts as Observable State

PKOS preserves observable state through structured reasoning artifacts. Instead of capturing only outputs, a reasoning artifact records elements of the reasoning process itself:

With these preserved, a reasoning system keeps an observable record of how its conclusions emerged.

Feedback Loops in Reasoning Systems

When reasoning artifacts remain visible, reasoning systems form the kind of feedback loop cybernetics studies:

reasoning → record → evaluation → correction → continuation

Such a loop lets reasoning be revisited and revised instead of disappearing once a conclusion is reached. It is what supports cumulative reasoning: the ability of a system to extend prior understanding across time.

Regulation Without a Controller

The systems view's least intuitive result is that regulation does not need a central controller. A road network is regulated by a loop of sensors, signals, agents, and feedback, with nobody in charge of it. The same structure is available for reasoning, in which the loop does the regulating and no single institution has to see everything.

→ Systems Perspective, the same ideas as traffic

Human Oversight and Adaptive Systems

Cybernetic systems include a part that interprets and adjusts. In human reasoning environments that part is fundamentally human. Remove it and what remains is automated regulation that no one can interpret.

PKOS keeps the human in the loop at a defined point: participants interpret reasoning artifacts, evaluate conclusions, and take responsibility for what becomes a durable commitment, at the Promotion Gate. That is what keeps an adaptive reasoning system interpretable and accountable instead of opaque.

Reasoning Infrastructure as a Learning System

Seen this way, reasoning infrastructure is a learning system. When reasoning artifacts remain observable and revisable, a system can refine its interpretations, correct its errors, and extend understanding across cycles of inquiry, which is the cybernetic principle that learning emerges when a system can observe, evaluate, and adjust its own behaviour.

Where This Stands

This is independent research. The loops and checkpoints described here exist as working infrastructure. What they have not yet had is validation against real systems in real conditions: whether they improve the regulation of reasoning is the empirical question this programme wants tested, with partners. The most useful thing anyone from this field could tell us is where the mapping holds and where it breaks.

→ Contact
→ The sibling question, from the side of meaning and signal: Information Theory

Cybernetics was built to answer a question like the one now facing reasoning systems: how does something stay on course when conditions keep changing and no one can see everything? The field has decades of answers. What is new is the thing being steered.

Related Concepts