101 – We are all living inside models

There is an experiment that every reader of this article can conduct right now, without equipment or preparation. Hold this page at normal reading distance and focus on the word at the center of the line you are currently reading. Without moving your eyes, try to make out the words at the far left and far right margins. They are visible — they are within the field of vision — but they are blurred, uncertain, recognizable only as shapes rather than letters. This is not an optical defect. The eye’s capacity for sharp resolution is, in fact, confined to a surprisingly small region at the center of the visual field. The sense that we see the whole page clearly, all at once, is an illusion — a construction that the brain maintains by rapidly moving the eye across the page and stitching together the results into a seamless impression of uniform clarity.1

This is a small example of a very large phenomenon. What we experience as perception — the immediate, apparently direct apprehension of the world around us — is not perception at all, in any straightforward sense. It is interpretation. It is the end product of a process of inference, construction, and filling-in that the brain performs continuously and unconsciously, drawing on prior experience, current expectation, and the statistical regularities of a world it has learned to anticipate. We do not see the world and then build a model of it. We run a model of it, continuously, and what we see is the output of that model.2

This series is about that model. It is about all the models, in fact — the dozens of overlapping frameworks, narratives, and representations that each of us carries and that, together, constitute what we take to be our understanding of the world. It is about where those models come from, what they are good for, where they fail, and what it would mean to hold them with somewhat more honesty than we customarily manage.

The starting point: a finding, not a provocation

The claim that we do not have direct access to reality sounds, at first encounter, like the kind of statement that philosophers make to start arguments at dinner parties — interesting to debate, irrelevant to practical life. It is neither of those things. It is a finding — a result produced by every serious line of inquiry into the nature of perception, cognition, and neuroscience over the past hundred and fifty years. It has direct consequences for how we should think about our own confidence, our disagreements with others, and the nature of the knowledge we act on every day.

Consider a familiar situation. Two people watch the same political debate. They are in the same room, hearing the same words, observing the same expressions, absorbing the same information. Afterwards, they reach opposite conclusions — not about minor details of policy, but about who was honest, who was evasive, who had the better argument, and what the exchange revealed about the character of the candidates. We tend to explain this divergence by invoking bias, tribalism, or motivated reasoning — as though one person was thinking clearly and the other had allowed their emotions to distort their judgment.

This explanation is almost certainly wrong, or at least profoundly incomplete. Both observers were running models. They brought different prior beliefs, different histories of experience with the people involved, different frameworks for reading human behavior and political language — and those frameworks determined, before a single word of the debate was spoken, what each of them was capable of noticing. The debate did not produce their conclusions. Their models did. The question of which model was more accurate is a legitimate question and one worth asking carefully. But it is a different question from the one we usually ask, which is which observer was thinking and which was feeling. Both were thinking. Both were filtering.

The neuroscientist Karl Friston has formalized this observation into a comprehensive theory of how the brain works.3 In Friston’s framework, the brain is a prediction machine: it generates a model of the current state of the world, uses that model to predict incoming sensory data, and then updates the model when the data fails to match the prediction. The updating is proportional to the size of the mismatch — a small discrepancy is absorbed with minor revision; a large one triggers a more substantial restructuring. What we experience as perception is, on this account, the brain’s best current guess about the state of the world, constrained by but not simply read off from the sensory evidence.

The theory makes sense of phenomena that the simpler picture — the brain as a camera that records what is in front of it — cannot explain. Optical illusions are not failures of the visual system; they are demonstrations of its normal operation. The Muller-Lyer illusion, in which two lines of equal length appear different because of the direction of the arrowheads at their ends, persists even when we know the lines are equal and have measured them with a ruler. The model overrides the evidence. This is not a bug; it is a feature — a consequence of the fact that the brain has learned, from a lifetime of experience, that the configuration of lines with outward-pointing arrowheads typically corresponds to a more distant object and is therefore physically longer. The model that generates the illusion is, in the general case, more reliable than raw retinal data. It fails, specifically and instructively, in the artificial conditions of the psychologist’s laboratory.

What we mean by a model

The word model is used in so many contexts — mathematical models, role models, model citizens, fashion models — that its technical meaning has been almost entirely obscured. In this series, it carries a specific meaning that is worth establishing clearly at the outset.

A model, as the term is used throughout these articles, is any internal representation that organizes experience, generates expectations, and guides action. By this definition, almost everything that can be called knowledge is, in some sense, a model. Our understanding of gravity is a model: a representation of the relationship between mass, distance, and attractive force that allows us to predict where a thrown stone will land. Our understanding of a close friend’s character is a model: a representation of their likely responses to various situations that allows us to predict how they will react to news, pressure, or provocation. Our understanding of our own personality is a model: a representation of our characteristic tendencies, values, and capacities that guides our choices about what to attempt and what to avoid.

These models differ along several dimensions. Some are explicit and formalized — the equations of classical mechanics, the diagnostic criteria of the DSM, the provisions of a legal code. Others are tacit and embodied — the skilled cyclist’s model of balance, the experienced teacher’s model of a disengaged student, the musician’s model of where the next note should go. Some can be articulated in language or mathematics; others exist only as dispositions to act in particular ways and cannot be stated but only demonstrated. What they share is the functional role: they take the incoming stream of experience, organize it into patterns, generate predictions about what will come next, and guide behavior in response to those predictions.

Three classes of models deserve particular attention in the context of this series, because they operate at different scales of abstraction and with different degrees of difficulty in the task of examination.

The first class consists of models of non-living physical systems — the domain of the natural sciences. These are the models that describe the behavior of particles, fields, organisms, and ecosystems. They range from the equations of quantum electrodynamics, which predict the behavior of subatomic particles to twelve decimal places of accuracy, to the climate models that project global temperatures under various emissions scenarios, to the epidemiological models that estimate the trajectory of infectious disease. What distinguishes this class of models is the precision of the language in which they are expressed and the rigor of the methods by which they are tested. Mathematics, as a later article in this series argues, is the most successful modeling language humanity has ever invented — a formal system whose rules of inference are unambiguous and whose predictions can, in principle, be checked against observation with arbitrary precision.4

The second class consists of models of other people — the domain of what philosophers call folk psychology.5 These are the models we use to predict, explain, and influence the behavior of the human beings around us. When we attribute beliefs, desires, intentions, and emotions to other people — when we say that a colleague is annoyed because she believes her contribution was not acknowledged, or that a friend will enjoy a particular film because he values moral complexity — we are deploying a model. This model is extraordinarily sophisticated: human beings are, on any reasonable assessment, the most complex objects in the known universe, and our ability to navigate social life with reasonable success depends on maintaining workable representations of dozens or hundreds of other minds simultaneously. But it is also systematically imprecise in ways that mathematical models are not, and its errors — while less dramatic than a collapsed bridge or a failed rocket — compound across a lifetime of social interaction with consequences that are easy to underestimate.

The third class consists of models of ourselves — the representations we maintain of our own character, capacities, motivations, and histories. This class is the most interesting and the most difficult to examine, for a reason that should be immediately apparent: the instrument we use to examine our self-model is the same instrument whose accuracy is in question. The self is both the subject and the object of The Conscious Look, and the circularity is not an obstacle to be overcome but a condition to be acknowledged.

Why models are always wrong

Korzybski’s aphorism — the map is not the territory — is now so familiar that it has nearly lost its force. It is worth pausing on what it actually claims, because the claim is more radical than the metaphor suggests.

A map is wrong about the territory in at least three ways. It omits most of the territory’s features, representing only those relevant to the purpose for which the map was made. It simplifies the features it does represent, abstracting continuous variation into discrete categories, replacing curved lines with straight ones, substituting symbols for the complex textures of actual ground. And it introduces errors — small distortions, outdated information, features that have changed since the map was drawn — that accumulate in proportion to the scale and age of the map.

These three forms of inadequacy — omission, simplification, and error — characterize every model in every domain, without exception. A model that included all the features of the territory it represented would not be a model; it would be the territory itself. A map of England at the scale of 1:1 would be as useless as no map at all — it would have to be folded up in order to use it, and then it would be the same size as England and indistinguishable from it. The utility of maps, and of models generally, is a direct consequence of their incompleteness. They work because they omit. What they omit is what creates their limits.6

The practical consequence is that every model has a domain of validity — a range of conditions within which it provides useful guidance, and outside which it fails in ways that may or may not be visible from within the model itself. Newton’s laws of motion are a model of exceptional power and precision within the domain for which they were developed: bodies of moderate size moving at speeds well below that of light. At very high speeds, they fail. At quantum scales, they become inapplicable. But they give no internal warning of these limits — they do not become gradually less accurate as conditions approach the boundary. They are precise and reliable right up to the edge of their validity, and then they are simply wrong.

Most of the models we use in everyday life are less formally specified than Newton’s laws, which means their limits are harder to identify and easier to overstep. The model of human motivation that makes someone a skilled manager within a familiar cultural context may produce predictable failures when applied in an unfamiliar one. The political model that correctly identifies the causes of one kind of social dysfunction may generate disastrously wrong prescriptions when applied to a different kind. The self-model that correctly captures the person one was at thirty may constrain the choices of the person one is at fifty. In each case, the model is being used outside its domain of validity — often without any awareness that a boundary has been crossed.

The through-line of this series

The articles that follow this introduction explore the implications of the model-building account of mind across a wide range of domains. They move from individual cognition to social interaction to scientific method to political philosophy; from the nature of perception to the structure of scientific revolutions to the ethics of certainty; from the question of what intelligence is to the question of what artificial intelligence cannot be.

These articles are not primarily about the content of any particular model — not, that is, primarily about what the correct views on politics, science, or personal conduct are. They are about the form of model-building itself: the cognitive mechanisms that produce models, the social and institutional processes that transmit and maintain them, the systematic biases that distort them in predictable directions, and the practices that can at least partially correct those distortions.

The argument that runs through all of them is a modest one. It does not claim that all models are equally bad, or that the impossibility of direct access to reality means that no position is better supported than any other. It does not counsel paralysis or endless revision. What it argues is that the model is not the territory — that the gap between our representations and the reality they represent is real, consequential, and never fully closed — and that this fact, taken seriously, should modify how we hold our beliefs, how we respond to disagreement, and how we present our conclusions to others.

The modification is not radical. It does not require abandoning convictions. Most of us are, most of the time, justified in acting on the best model we have rather than suspending judgment until some impossible standard of certainty is reached. But there is a difference between acting on a model with appropriate confidence and mistaking the model for the world. The first is the only rational response to being finite, social creatures in a complex environment. The second is a cognitive error with consequences that range, depending on the stakes, from minor inconvenience to catastrophe.

The distinction between the two is not, in the end, a matter of intelligence or education — although both help. It is a matter of habit: the habit of occasionally turning one’s attention back on the frameworks through which one sees, and asking whether they still serve. This is what we mean by The Conscious Look.

The name

The title deserves a word of explanation, because it names a practice rather than a position.

A conscious look is not a permanent state of radical skepticism, in which every belief is held provisionally and every action is preceded by an epistemological audit. That is not a description of careful thinking; it is a description of paralysis, and it would make the conduct of ordinary life impossible. Most of what we do, we do on automatic — running established models in established conditions, trusting the outputs because the outputs have generally been reliable, reserving deliberate scrutiny for the situations where something seems wrong or where the stakes are unusually high.

The Conscious Look is something narrower and more feasible than permanent skepticism. It is the practice of periodically — not constantly, but deliberately — directing attention to the models themselves rather than through them. It is the habit of asking, with genuine openness, the questions that the normal conduct of thought leaves unasked: where does this belief come from? What evidence would change it? Where are its limits, and do I know what they are? What would I have to give up — socially, emotionally, professionally — if it turned out to be wrong?

These are not comfortable questions. They are the questions that make us reluctant to examine our most important convictions — the ones that, in a phrase that captures something real about political life, are strong enough to send people into the street. Those convictions are not wrong simply because we hold them strongly. But they are the ones most in need of examination, precisely because the social and emotional costs of revision are highest, and those costs exert a systematic pressure on the mind to find reasons not to look.

The articles in this series are, each in their own way, invitations to look — at the models of perception and cognition that generate our experience, at the models of science and language that structure our collective knowledge, at the models of the self and of other people that shape our relationships, and at the models of society and politics that organize our public life. The looking is uncomfortable. It is also, we have found, considerably more interesting than the alternative.

Notes

The technical term for this property of the eye is foveal vision. The fovea is a small region at the center of the retina with a very high density of cone photoreceptors, responsible for sharp color vision. Outside the fovea, resolution drops rapidly. The brain compensates by constructing a stable, detailed visual scene from a series of rapid eye movements called saccades — typically three to four per second — each of which brings a different part of the scene into foveal focus. The seamless impression of uniform clarity is a post-hoc construction, not a direct record.
The technical term for this class of theory is predictive processing, sometimes also called the predictive coding framework. The most influential formulation is due to the neuroscientist Karl Friston, building on earlier work by Hermann von Helmholtz in the nineteenth century. Friston’s free energy principle proposes that minimizing prediction error — the mismatch between the brain’s model of the world and incoming sensory signals — is the fundamental objective of neural processing. The theory is controversial in its strong forms but has generated substantial empirical support in its core claims about the role of prior expectations in perception.
Friston, K. (2010). The free-energy principle: a unified brain theory? Nature Reviews Neuroscience, 11(2), 127-138. For a more accessible treatment, see Clark, A. (2016). Surfing Uncertainty: Prediction, Action, and the Embodied Mind. Oxford University Press.
The philosopher Eugene Wigner called this the unreasonable effectiveness of mathematics in the natural sciences — the puzzling fact that mathematical structures developed for purely internal reasons of elegance and consistency turn out to describe physical reality with extraordinary accuracy. The puzzle is discussed at length in the mathematics article (Article 6) in this series.
Folk psychology is the term philosophers use for the everyday practice of explaining and predicting human behavior by attributing mental states — beliefs, desires, intentions, emotions — to other people. It is called folk not because it is unsophisticated but because it is the psychology practiced by all human beings without formal instruction, in contrast to scientific psychology. Whether folk psychology is a good theory of mind, a useful fiction, or something else entirely is one of the central debates in the philosophy of mind.
The observation is due to Alfred Korzybski, who introduced the phrase the map is not the territory in a 1931 address to the American Mathematical Society. Korzybski’s broader project — the development of what he called General Semantics, a discipline aimed at clarifying the relationship between language, thought, and reality — was influential in the mid-twentieth century and deserves more attention than it currently receives. His central work is Science and Sanity (1933), which is demanding but rewarding.