A government never encounters the society it governs in its entirety. A health authority does not perceive every illness, behaviour, household circumstance and interaction occurring across a population at once. A transport agency does not experience a city as millions of simultaneous journeys, changing intentions, informal routes, delays, economic activities and individual decisions. A ministry responsible for employment cannot hold every worker, employer, skill, vacancy, contract, aspiration and local labour-market condition inside a single act of judgement. Reality arrives at institutions selectively, through categories, measurements, reports, cases, professional observations and administrative systems. Before an institution can govern the world, it must therefore possess some organised representation of what that world is like.
That representation is a model of reality. The word “model” can suggest something formal or technical — an economic forecast, a simulation, a statistical model or an artificial-intelligence system — but institutional models are much broader than these explicit analytical tools. A policy framework contains a model of the problem it addresses. An organisational structure contains assumptions about how responsibilities relate. An eligibility system embodies a representation of relevant differences between people. A performance dashboard expresses a view of which properties of a service matter sufficiently to observe. Even an ordinary administrative form contains a miniature model of reality because its questions determine which characteristics of a person or situation the institution needs to recognise.
Institutions need such models because reality contains far more information than any organisation could use simultaneously. Governing requires selection. If a city wants to manage traffic, it does not need a representation containing every fact about every person currently within its boundaries. It needs information about movements, capacities, bottlenecks, routes, demand and other properties relevant to the transport problem. A useful model reduces the world to a structure that allows institutional attention and action to become possible.
This means that omission is not automatically a weakness. Indeed, the usefulness of a model often depends on what it leaves out. A map that attempted to reproduce every physical detail of a city would be less useful for navigating its underground network than a diagram containing only stations, lines and connections. The omitted streets, buildings and topography are real, but they are not equally relevant to that particular task. The map succeeds partly because it excludes them. Institutional models operate through the same principle: productive simplification allows complexity to become cognitively and operationally manageable.
The question is therefore not whether a model omits reality. Every model does. The more important question is whether it preserves the distinctions that matter for the purpose for which it is being used. An unemployment model that distinguishes only between employed and unemployed people may be sufficient for one statistical purpose while being inadequate for designing support for people moving repeatedly between temporary work, informal employment and periods outside the labour market. Neither judgement can be made simply by asking whether the model is detailed. What matters is whether its simplifications correspond to the institutional question being asked.
This is why a more complete model is not necessarily a better model. Additional variables, categories and relationships can increase representational richness, but they can also make a model harder to interpret, maintain and use. Institutions need representations capable of reducing complexity without destroying the properties necessary for intelligent action. There is no general requirement to represent everything, and in many contexts attempting to do so would be both impossible and undesirable.
The impossibility matters. Social systems contain vast numbers of interacting actors, histories, preferences, incentives and relationships, many of which change continuously. No administrative system could reproduce that complexity in full. Even if an institution could collect dramatically more information, the resulting representation would still involve choices about categories, relationships, time horizons and relevance. More data can make a model richer, but data accumulation does not remove the need to model.
The undesirability matters too. Institutions are not entitled to know everything merely because information might improve a representation. Privacy, autonomy, proportionality and democratic limits constrain what governments should observe. A perfectly detailed representation of citizens’ lives would not simply be technically difficult; in many circumstances it would be institutionally unacceptable. Good governance can therefore require deliberate ignorance as well as deliberate knowledge: an institution may properly exclude information because possessing it would be unnecessary, intrusive or inconsistent with the limits of public authority.
This creates an important distinction between representational incompleteness and representational failure. Every institutional model is incomplete. Failure occurs when the omitted or distorted features are sufficiently important to the task that the institution’s resulting understanding becomes unreliable. A transport model that ignores the colour of travellers’ clothing is incomplete but probably not defective. A model of accessibility that ignores whether people with disabilities can actually use the transport network may be incomplete in a way that directly compromises the purpose for which the representation is needed.
Models therefore cannot be judged independently of purpose. The same representation may be highly effective for one decision and dangerously inadequate for another. Administrative categories created to determine eligibility for a benefit may work well for processing claims while providing a poor representation of social vulnerability. A budget model designed to allocate annual expenditure may be excellent at controlling short-term commitments and weak at representing long-term resilience. A model of hospital performance built around throughput may support capacity management while obscuring dimensions of patient experience that matter for evaluating care.
Because models organise what institutions can see, they also influence what institutions can think about systematically. Once a phenomenon has a recognised category, indicator or relationship inside an institutional model, it becomes easier to compare, discuss and connect to decisions. Features outside the model may still be known informally, but they have less structured access to institutional attention. Models are therefore not passive mirrors of reality. They participate in organising the cognitive environment within which institutional reasoning occurs.
This influence can be seen whenever different models of the same situation produce different policy possibilities. A neighbourhood represented primarily through crime statistics invites one set of questions; represented through housing conditions, mobility, social networks and access to services, it invites others. Neither representation necessarily contains false information. The difference lies in what relationships each model makes salient. The institution’s picture of the problem affects the interventions it can readily imagine.
Yet recognising this power should not lead to the opposite mistake of treating all models as arbitrary. Reality constrains representation. Some models explain observations better than others, make more reliable predictions, preserve more relevant distinctions or support more effective decisions. Institutional models can be tested against experience and challenged by evidence. The fact that they are constructed does not mean that every construction is equally useful.
What matters is that the institution retains the distinction between the model and the world the model represents. A model is an instrument for reasoning about reality, not reality itself. Its categories may be useful without being natural divisions of the world. Its indicators may reveal important properties without exhausting everything that matters. Its causal assumptions may guide action while remaining open to revision. Its boundaries may make governance possible without defining the true boundaries of the underlying problem.
This distinction becomes particularly consequential when a model works well for a long time. Success can make its assumptions less visible. Categories become embedded in databases, regulations and organisational structures; indicators become familiar; relationships that were once hypotheses become routine premises. The representation can gradually acquire the appearance of inevitability because the institution has organised so much of its activity around it. The model remains a model, even when the organisation no longer experiences it as one.
Artificial intelligence makes the distinction no less important. Machine-learning systems can construct representations containing relationships too numerous or complex for people to hold directly, and this may allow institutions to anticipate outcomes with remarkable accuracy. But sophistication does not eliminate modelling. Decisions about training data, variables, objectives, categories and optimisation still determine which aspects of reality enter the system and how they are represented. A model containing millions of parameters remains selective in relation to the world from which those parameters were derived.
The same principle applies to simpler institutional artefacts. A spreadsheet, risk register, organisational chart or case-management system can shape institutional cognition precisely because it embodies a model of what needs to be represented. Some models are explicit and mathematically sophisticated; others are embedded so deeply in administrative routines that they are barely noticed. Their cognitive importance does not depend on their technical complexity. It depends on how strongly they structure what the institution can recognise and act upon.
A cognitively capable institution therefore does not aspire to eliminate models or replace them with direct access to reality. Such access does not exist. Instead, it develops the capacity to know what its models are for, which distinctions they preserve, which features they intentionally omit and where their limits become consequential. It can use simplification without confusing simplicity with completeness, and it can recognise that a model may remain useful even while acknowledging what lies outside it.
This form of model awareness is not an argument for permanent hesitation. Institutions must act, and action often requires treating a working representation as sufficiently reliable for the decision at hand. The important discipline is to preserve the difference between saying, “this model is good enough for this purpose” and saying, “this is what reality is”. The first is a defensible institutional judgement. The second collapses the distinction on which intelligent modelling depends.
Every institution governs through a model of reality because governing reality directly is impossible. Models make action possible by selecting, simplifying and organising the complexity of the world into forms that institutions can perceive and use. Their incompleteness is therefore not automatically a defect; productive omission is part of what makes a model useful, and total representation would often be impossible, undesirable or both. Institutional intelligence depends on constructing models that preserve what matters for the task while remaining aware of what they leave out. The model must be good enough to guide action, but it must never be mistaken for the reality it was built to represent.