Research
DCFB: Distributed Cognition as Foundational Behavior: Oscillatory Fields
A theoretical framework for understanding how intelligence distributes across human-AI systems: and why the unit of cognitive analysis must shift from the individual to the field.
Consider a modern hospital. Every doctor is brilliant. Every nurse is highly trained and experienced. The administrative staff is diligent. The diagnostic machines are state of the art.
Yet, somehow, a patient’s chart gets lost, a medication allergy is overlooked, and a scheduled surgery is delayed by six hours.
If you try to locate the “stupidity” in this system, you will fail. No individual person acted stupidly. No single machine malfunctioned. The intelligence of the people was high. But the intelligence of the hospital was low. The failure did not occur inside anyone’s head; it occurred in the space between the heads. It occurred in the handoffs, the poorly designed forms, the conflicting software systems, and the unwritten rules about who is allowed to interrupt whom.
When we talk about intelligence, we have a cultural habit of locating it inside a single container. We point to the intelligent person, the intelligent algorithm, or the intelligent model.
This model of located intelligence breaks down immediately when you try to understand how complex decisions actually get made. If a CEO makes a major strategic pivot, the decision was not made by the CEO alone. The decision was made by the CEO, her executive team, the financial analysts who built the forecasting model she consulted, the consultants who framed the strategic options, the board members whose concerns shaped the framing, and the AI assistant that surfaced the key counterexamples.
The intelligence was distributed across a field of actors, tools, and structures. If you want to govern that decision, looking only at the CEO’s brain is useless. You have to look at the field.
This brings us to DCFB: Distributed Cognition as Foundational Behavior. DCFB takes this insight seriously as the absolute starting premise for institutional design. It argues that intelligence is not located in any one node of the system. It emerges from the field.
The Field Boundary
If cognition is distributed, then the most important concept in governance is not the capability of the agent, but the location of the Distributed Cognition Field Boundary (the DCFB itself).
The field boundary is the invisible line where the system’s memory, coordination, and shared reality actually stop. This boundary rarely matches the official organizational chart. An engineering team might be officially part of a division, but their actual cognitive field boundary might include an open-source community they rely on, an AI code assistant they use daily, and a slack channel that includes people from three different departments.
If you are an executive trying to govern that engineering team, and you only govern the nodes inside the official organizational chart, you are blind to the actual cognitive system doing the work. You have an asset layer gap: a disconnect between the assets you think you are governing and the actual cognitive assets that are producing the outcomes.
The Principle at Three Scales
Shifting the unit of analysis from the individual node to the distributed field changes how we evaluate capability at every scale.
At the individual scale. A commercial airline pilot flying through a storm is not relying solely on the intelligence in their own head. They are part of a cognitive field that includes the cockpit instruments, the autopilot algorithms, the air traffic controller on the ground, and the physical design of the yoke. The pilot’s ability to navigate the storm is not a measure of individual genius; it is a measure of how well the pilot is integrated into that specific cognitive architecture. If you replace the familiar instruments with completely new ones mid-flight, the pilot’s internal intelligence does not change, but the field’s intelligence drops to near zero.
At the team scale. A surgical team in an operating room often coordinates complex procedures without speaking a word. The scrub nurse anticipates the surgeon’s need for a scalpel based on the angle of the surgeon’s hand and the stage of the procedure displayed on the monitor. The cognition required to perform the surgery is distributed across the entire team, the instruments, and the layout of the room. If you swap out the scrub nurse for an equally skilled nurse who has never worked with this surgeon, the surgery slows down. The individual nodes are just as smart, but the relational connections within the field have been severed.
At the organizational scale. An enterprise company decides to adopt a massive new AI infrastructure. The standard approach is to treat the AI as a new “node” and simply plug it into the existing organizational chart. Management assumes the organization will instantly become more intelligent. But under DCFB, the organization realizes that plugging a highly capable node into a broken field does not fix the field; it just accelerates the field’s existing dysfunctions. The organization must redesign the cognitive field itself. They must rebuild how information flows, how authority is delegated to the AI, and how humans monitor the boundaries of the AI’s actions.
A highly capable model embedded in an organization with weak cognitive infrastructure is less intelligent, effectively, than a less capable model embedded in an organization with strong cognitive infrastructure.
What Changes When You Treat Cognition as Distributed
When you accept that intelligence is a field phenomenon rather than a node phenomenon, the entire paradigm of AI governance shifts.
The unit of analysis shifts. You stop asking “how intelligent is this AI model?” and start asking “how intelligent is this system as part of our specific operational field?” The focus moves from benchmark scores in a vacuum to integration quality in reality.
Governance requirements change. If intelligence is distributed, then accountability must be distributed in ways that match the actual distribution. You cannot locate legal or moral accountability solely in the person who pressed the final “approve” button if the decision that mattered was formulated, shaped, and finalized by an AI agent three steps earlier in the workflow.
The design problem changes. Designing for distributed cognition means designing the relationships between nodes, not just the capability of any single node. The cognitive architecture (the structural arrangement of human and machine intelligence) matters more than the raw computing power of any element within it.
Failure modes become visible in different places. Failures in distributed cognitive systems rarely appear as a single broken component. They appear in the interfaces between nodes. The handoffs. The assumptions made by one system about the data formatted by another. The failure is distributed too.
The Five Core Principles of DCFB
The DCFB framework is built on five foundational principles.
Principle 1: Intelligence is field-emergent, not node-located. Any assessment of cognitive capability must include the entire field in which the capability operates. An AI system that scores flawlessly on an academic exam might fail catastrophically when embedded in a high-stress corporate supply chain, because the field dynamics alter the expression of the intelligence.
Principle 2: Fields have constitutions, explicit or implicit. Every cognitive field operates according to rules about authority, information flow, and accountability. These rules may be written in a policy manual, or they may be unwritten cultural habits. But they are always present, and they always govern the field’s behavior.
Principle 3: Constitutional design precedes capability optimization. Before you try to optimize the raw capability of any node in the field, you need to deliberately design the field’s constitution. Capability without a constitutional structure creates severe fragility. Building a hyper-intelligent AI agent without defining who it reports to and how it is halted is building a bomb, not a tool.
Principle 4: Distributed accountability requires distributed monitoring. You cannot monitor a distributed cognitive system from a single, centralized dashboard. Accountability structures must be distributed throughout the network in ways that match the actual distribution of intelligence and action.
Principle 5: Cognitive fields are dynamic, not static. The field changes as new nodes are added, as relationships between human workers and AI agents evolve, and as the external market environment shifts. Governance cannot be a static document; it must be an adaptive, ongoing process.
The Ghost Lineages and the Two-Notebook Architecture
The DCFB framework is not just a theoretical model; it is deeply rooted in the evolutionary history of how intelligence develops.
The Notion insight logs (specifically IL-016 from March 2026) document the “Ghost Lineages Thread.” This entry reveals the evolutionary substrate of DCFB. It notes that biological intelligence did not evolve simply by building larger individual brains. It evolved through altitude adaptation and cognitive architecture that distributed processing across living systems. The ‘eigenform’ (the core structural pattern) of intelligence persisted through distribution into the ecology, not through isolated archival storage. Biological cognition has always been distributed. We are simply waking up to the fact that artificial cognition must be distributed too.
Furthermore, IL-014 documents the “Two-Notebook Architecture,” revealing a critical tension in how DCFB was formulated. The log notes that there is a difference between the public presentation of institutional authority and the private, phenomenological reality of how the insights were generated. The framework recognizes that you cannot fully govern a system if the “phenomenological substrate” (the actual, messy reality of how the field operates) is hidden behind a polished institutional facade. True distributed cognition requires visibility into the actual field mechanics, not just the sanitized reports.
Connection to the CIR Framework
The Cognitive Infrastructure Readiness (CIR) index is the tool that operationalizes the DCFB theory for enterprise deployment. The five constitutional dimensions of the CIR readiness assessment map directly onto the five DCFB principles:
- Intent Specification maps to Principle 1 (field-level capability assessment). You must specify intent for the whole field, not just one node.
- Authority Architecture maps to Principle 2 (explicit constitution). You must document the actual rules of authority.
- Failure Mode Literacy maps to Principle 3 (constitution before optimization). You must understand how the field fails before making it run faster.
- Alignment Monitoring maps to Principle 4 (distributed monitoring). You must track accountability across the network.
- Governance Scalability maps to Principle 5 (dynamic governance). You must build systems that adapt as the field changes.
DCFB is the theoretical foundation. CIR is the practical instrument.
DCFB Governance Primitives
To move from theory to execution, the DCFB framework defines a set of Governance Primitives. These are the fundamental rules that must be embedded into the actual source code or system prompts of any agentic AI system operating within the field.
- Recursive Sovereignty: The capacity for any node (or sub-agent) to maintain local autonomy over its internal state, provided its external outputs remain calibrated to the parent field’s constitution. Sovereignty is not a binary switch; it is a recursive property of the entire hierarchy.
- Substrate Agnostic Execution: Cognitive integrity must be maintained regardless of the underlying hardware or software substrate. Whether the node is a human brain, a cloud-based Python script, or a local JavaScript runtime, the rules of the field apply. The Field is the invariant; the Node is the variable.
- Agentic Accountability: In a distributed field, accountability is moved from the individual actuator to the specific Authorization Layer that enabled the action. If an AI sends a damaging email, we do not blame the language model; we audit the authorization path that permitted the send command to execute without a human gate.
- Structural Connectedness: The intelligence of a field is directly proportional to its structural connectedness. A field with high connectivity but low calibration is noisy. A field with high calibration but low connectivity is brittle. Optimal intelligence requires a calibrated resonance across all relational paths in the network.
If you want to build a truly intelligent system, stop staring at the nodes. Map the field.