When institutions try to improve how a group thinks, they often begin with the people. They recruit different expertise, provide training, change leadership, add analytical capacity or bring new participants into the decision process. These interventions can matter enormously because the knowledge, judgement and experience available to a system obviously depend partly on who participates in it. Yet there is another source of cognitive capability that is easier to overlook. The same people, possessing the same individual knowledge, can produce very different collective understanding depending on how they are connected to one another.
Imagine a group of public officials distributed across several units. Each person knows something that the others do not. In one configuration, information flows upward through separate hierarchies and only meets at senior level. In another, specialists can communicate directly across organisational boundaries. In a third, one coordinating unit becomes the principal intermediary through which most exchanges occur. Nothing about the individuals has changed, yet the cognitive properties of the system can change substantially. Some configurations allow distant pieces of knowledge to meet quickly; others keep them apart. Some expose disagreement early; others filter it through several layers. Some create multiple routes through which weak signals can travel; others depend on a small number of cognitive bottlenecks.
This gives institutional cognition a relational dimension. SAME PARTICIPANTS + DIFFERENT RELATIONSHIPS = DIFFERENT COLLECTIVE COGNITION. Who is present matters, but so does who can observe whom, who can question whom, where information converges, which actors mediate between otherwise separated groups and how easily a local insight can become consequential elsewhere.
A network is therefore more than a communication infrastructure. Communication channels matter, but relational architecture also shapes attention, interpretation and influence. If two units exchange reports but neither can challenge the assumptions through which the other interprets them, information may circulate without much cognitive integration. Conversely, a relatively modest connection between previously separated specialists can sometimes transform institutional understanding because it allows two complementary representations of a problem to interact directly.
This is why CONNECTION ≠ COGNITIVE RELATIONSHIP. An organisation chart may show formal links, digital platforms may allow universal messaging and meetings may bring many actors together, yet the effective network through which institutional cognition occurs can remain very different. Some relationships carry authority, some trust, some specialised knowledge, some contextual interpretation and some access to decision-makers. Treating every connection as cognitively equivalent obscures the architecture that actually shapes collective thought.
Network position matters for the same reason. An actor located between otherwise weakly connected communities may possess no greater individual expertise than colleagues elsewhere, but its position can give it unusual cognitive importance. It can encounter several interpretations of the same problem, detect contradictions that remain invisible within specialised groups and translate knowledge across boundaries. Removing that relationship, or routing all exchanges through a different intermediary, can change what the wider institution is able to notice without changing the knowledge possessed by any individual member.
The opposite is also possible. Highly centralised networks can make coordination efficient because information converges rapidly on a small number of nodes. But those nodes can become cognitive bottlenecks. They must decide what deserves attention, translate material originating in different domains and redistribute what they consider relevant. Even competent intermediaries inevitably filter. As the volume and diversity of information increase, more of the system’s collective cognition becomes dependent on the capacity and assumptions of a small number of actors.
More decentralised networks can preserve multiple pathways through which knowledge travels, making the system less dependent on any single intermediary. They may also allow specialised communities to combine information directly. Yet additional connectivity does not automatically produce better cognition. If every participant communicates continuously with every other participant, attention can become overwhelmed, weak signals can be lost in informational volume and the cost of maintaining relationships can exceed their cognitive value. MORE CONNECTIONS ≠ MORE INTELLIGENCE.
The relevant question is therefore not how connected an institution is in the abstract, but what kinds of cognitive relationships its network makes possible. Can knowledge generated at the edge reach places where it changes judgement? Can actors holding different representations of a problem discover that their interpretations conflict? Can information bypass a bottleneck when the bottleneck itself is part of the problem? Can specialised communities remain sufficiently coherent to develop expertise while still maintaining bridges to knowledge outside their domain?
These properties matter because institutional problems rarely respect organisational boundaries. A policy failure observed by frontline staff may depend on a procurement decision made elsewhere, which in turn reflects a legal interpretation developed in another unit and a budget assumption established at the centre. If the relevant actors remain relationally separated, each can understand its own component while nobody sees the interaction. The institution does not necessarily lack knowledge. Its network prevents existing knowledge from becoming a wider explanation.
Changing the network can consequently change what becomes visible. A new cross-functional relationship may reveal that problems previously classified as local exceptions share a common cause. A direct connection between analysts and frontline teams may expose assumptions embedded in a model that looked reasonable from aggregated data. A relationship between policy designers and implementation units may allow operational consequences to alter policy reasoning before formal evaluation occurs. In each case, the cognitive improvement arises not because new knowledge has necessarily entered the institution, but because existing knowledge can now encounter other knowledge differently.
The same principle applies to disagreement. Networks determine not only how information moves but where conflicting interpretations meet. If disagreement must travel through hierarchical filters before reaching a common decision point, important differences may be softened, simplified or removed. If competing perspectives interact directly, the institution may be better able to identify whether disagreement arises from evidence, assumptions, professional frames or values. Relational architecture therefore influences the institution’s capacity to discover what kind of disagreement it actually has.
Trust adds another layer. Formal connectivity does not guarantee that actors will share uncertainty, admit failure or transmit information that challenges dominant expectations. Relationships characterised by low trust may carry routine data while suppressing precisely the knowledge most valuable for institutional learning. Conversely, strong relationships can allow weak, ambiguous or uncomfortable signals to travel before they have been converted into formally defensible claims. The cognitive network is therefore partly social as well as structural.
Artificial intelligence can modify this architecture in several ways. AI systems may connect previously separated bodies of information, identify relationships across departmental datasets or make specialised knowledge searchable by actors elsewhere in an organisation. This could reduce some forms of relational distance. Yet AI can also create new central cognitive nodes through which many institutional interpretations pass. If a common system becomes the principal mediator between distributed knowledge sources and decision-makers, its classifications, retrieval mechanisms and summarisation practices become part of the network architecture itself.
The important point is not that one topology is universally superior. Different cognitive tasks require different relational configurations. Rapid operational coordination may benefit from concentrated pathways, exploratory problem-solving from broader cross-boundary interaction, specialised reasoning from relatively dense professional communities and weak-signal detection from connections capable of bridging distant parts of the system. Institutional intelligence therefore cannot be reduced to maximising connectivity or decentralising every network.
Instead, relational architecture should be treated as a design variable. Institutions routinely redesign formal structures, reporting lines and responsibilities, but they can also ask how those changes alter the pathways through which cognition occurs. Closing a regional office, merging two departments, centralising an analytical function, creating an interdisciplinary team or introducing a shared digital platform does more than redistribute work. Each intervention can change which forms of knowledge encounter one another, where interpretation takes place and which actors become cognitively central.
This means that organisational reforms can have cognitive consequences even when cognition is not their explicit objective. A restructuring intended to reduce administrative duplication may remove a relationship that previously connected two knowledge communities. A coordination mechanism designed to simplify decision-making may create a new interpretive bottleneck. Conversely, a seemingly minor interface between units can produce disproportionate cognitive benefits if it connects perspectives that were previously isolated.
The institution therefore cannot understand its own intelligence by inventorying expertise alone. Knowing who possesses which knowledge is important, but it remains incomplete without understanding the relational architecture through which that knowledge can interact. Collective cognition is produced not simply by the contents of individual minds but by the configuration that determines which contents can meet, challenge, reinforce and transform one another.
Change the network and you change the thinking because cognition has a topology. The same participants, holding the same knowledge, can collectively notice different patterns, generate different interpretations and reach different judgements when the relationships among them change. An intelligent institution therefore pays attention not only to who knows what, but to how those who know different things are connected. Its cognitive capability resides partly in people and knowledge, but also in the architecture of relationships that determines what those people and that knowledge can become together.