No institution can perceive everything that matters to the world it governs. A ministry responsible for employment cannot know every individual working history, aspiration, informal arrangement, business decision and local economic condition at once. A health system cannot represent every aspect of every patient’s life while allocating resources across millions of cases. A city government cannot hold the full physical, social, economic and cultural reality of a city inside its administrative processes. The world is simply too detailed, too dynamic and too interconnected to enter an institution in its entirety. To govern at all, institutions must therefore reduce reality to something they can perceive and act upon.
They do this constantly. They divide populations into categories, territories into administrative units and problems into policy domains. They select indicators to represent performance, establish thresholds to distinguish ordinary from exceptional cases and build models that describe how different parts of a system relate. They use maps, statistics, classifications, eligibility rules, forecasts and reporting structures to transform an overwhelmingly complex environment into a world that can be administratively recognised. None of this is inherently a failure. Simplification is one of the conditions that makes organised action possible.
Imagine trying to govern unemployment without categories. Every person’s relationship with work is unique: employment may be stable or precarious, formal or informal, voluntary or involuntary, full-time or partial, interrupted by care responsibilities, education, illness or migration. An institution cannot design every policy around an unlimited description of every individual circumstance. It creates categories that make the population legible enough for programmes, budgets and responsibilities to be organised. The categories inevitably lose detail, but without some reduction of complexity there would be no administratively usable object called “unemployment” on which the institution could act.
The same principle applies to measurement. Governments cannot observe the economy directly as a complete living system. They observe selected properties through indicators: employment rates, inflation, productivity, tax receipts, business formation and many others. Each indicator makes part of reality visible while leaving other parts outside its frame. Taken together, these measures can create an extraordinarily useful representation of economic conditions. They remain a representation nonetheless.
This distinction between reality and its institutional representation is fundamental because representations shape action. What appears in an institutional model can become available for attention, comparison, budgeting and intervention. What does not appear may be much harder to recognise. If a service measures cases completed but not the effort citizens expend navigating the process, operational efficiency becomes highly visible while administrative burden remains comparatively obscure. If an education system represents performance primarily through test results, other dimensions of educational experience may receive less institutional attention even when everyone agrees they matter.
Visibility is therefore partly produced by the architecture through which an institution represents its environment. This does not mean that indicators fabricate reality or that what they measure is unreal. It means that measurement selects. A representation draws boundaries around what will count as an observable feature of the problem, and those boundaries influence what the institution is able to notice systematically.
Classification works in a similar way. Categories allow institutions to treat cases consistently, compare populations and allocate responsibilities. But every category groups together differences and separates cases that may share important similarities. A household can sit just above an eligibility threshold while being materially almost indistinguishable from one just below it. The threshold may still be necessary for administration. The cognitive danger begins only when the administrative distinction is mistaken for a complete description of the underlying social difference.
Simplification is therefore not the opposite of intelligence. Intelligent institutions simplify because they must. The more useful question is whether they understand what their simplifications preserve and what they remove. A map is useful precisely because it does not reproduce every feature of the territory. A transport map that included every building, tree and pavement would become less useful for navigating a railway network. Its power comes from selective representation. But using that map to understand housing inequality would be absurd because the features it excludes are essential to the new question.
Institutional representations work in much the same way. They are constructed for purposes, whether or not those purposes are always explicit. A classification designed to administer benefits may be poorly suited to understanding vulnerability. A performance indicator designed to monitor throughput may say little about quality. A financial model designed for annual budgeting may fail to represent long-term resilience. The fact that a representation works for one institutional task does not mean it is an adequate picture of reality for every other task.
This is why simplification is never entirely neutral. Selecting what to represent means deciding, explicitly or implicitly, which distinctions deserve institutional visibility. A model that divides a city by administrative districts produces one picture; a model organised around travel-to-work patterns produces another. Neither necessarily falsifies the city. They simplify it differently, and those simplifications make different relationships easier to see.
Problems arise when the representational choice disappears from view. Categories can become so familiar that they begin to look like natural properties of the world rather than administrative constructions. Indicators can acquire authority beyond the purpose for which they were designed. Organisational boundaries can make interconnected problems appear separate because responsibility has been divided between departments. Over time, the institution can begin to perceive reality through structures that were originally created merely to make reality manageable.
The consequences can be subtle. What fits the representation is easier to discuss because the institution has vocabulary, data and procedures for it. What falls outside may appear anecdotal, exceptional or difficult to justify. Staff can encounter a phenomenon repeatedly without possessing an authorised category through which to describe it. Citizens can experience problems that cut across several administrative systems while each system sees only the fragment corresponding to its own representation.
The issue is not that institutions should attempt to eliminate simplification. Such an ambition would make governance impossible. Even vastly better data and computing power cannot create a representation containing every potentially relevant property of a complex society. More detailed models simply make different choices about what complexity to retain. The challenge is therefore not to escape representation but to remain cognitively aware of its limits.
Artificial intelligence makes this particularly important. AI systems can construct representations from quantities of data far beyond ordinary human processing capacity, allowing institutions to identify relationships that simpler models would miss. Yet greater representational complexity does not abolish selection. Training data define what the system can learn from; target variables define what it is asked to recognise; classifications determine how outcomes are organised; optimisation criteria establish which distinctions matter to the system’s operation. A sophisticated model is still a model.
Indeed, greater analytical sophistication can sometimes make the distinction between representation and reality harder to remember. A model that performs extremely well may feel less like a simplified account and more like direct access to the world. But predictive accuracy does not transform a representation into reality. Every model remains conditioned by what it observes, how variables are constructed, which relationships it captures and the purposes against which its performance is assessed.
Institutional cognition therefore depends partly on representational self-awareness. An institution needs to know not only what its indicators, categories and models show, but also what kind of world those devices make visible. It needs to recognise that another representation might reveal different properties of the same underlying reality. This does not require permanent scepticism towards every administrative category. It requires the capacity to distinguish a useful simplification from the world that has been simplified.
Such awareness matters most when reality begins to change. A representation can remain internally coherent while the world moves beyond the assumptions that made it useful. Categories may persist after social practices evolve. Indicators may continue reporting stable performance while important outcomes migrate outside what they measure. Administrative boundaries may preserve an old division of responsibility even as problems become increasingly interconnected. If the institution treats its representation as reality itself, these changes become difficult to perceive.
The deepest cognitive risk is therefore not simplification but reification: forgetting that a constructed representation is constructed. Once that happens, what the institution cannot see through its existing categories may begin to look as though it does not exist. The representation ceases to function merely as a tool for navigating complexity and begins to define the limits of institutional reality.
Avoiding that risk does not mean governing without models. It means governing with models while remembering that models are selective. It means using indicators while understanding that indicators illuminate some dimensions more strongly than others. It means using classifications while recognising that boundaries created for administration may not correspond perfectly to boundaries in lived reality.
Every institution governs through a simplified world because reality is too complex to be represented in full. Categories, indicators, models, maps and administrative boundaries make collective action possible by reducing that complexity to forms an institution can perceive and use. The important question is therefore not whether institutions simplify, because they inevitably do, but whether they remain aware of what their simplifications make visible and what they leave outside. Institutional intelligence requires the ability to act through a representation without forgetting that the representation is not the world itself.