Judea Pearl begins with an act of humility that masquerades as a technical claim. The data, he tells us, cannot speak without a model. Before we can ask what causes what, before we can extract causal meaning from the flat record of correlations, we must first commit to a diagram — a picture of the world's structure, drawn by hand, placed over the data like a transparency over a map. The mathematics that follows this commitment is, as Pearl's admirers are quick to note, genuinely rigorous. Given the diagram, the inferences are valid. The do-calculus — the formal language Pearl developed for reasoning about interventions, for asking not just "what is correlated with what" but "what would happen if we changed this" — is a real and serious intellectual achievement.

But the diagram is not discovered. It is drawn. And the hand that draws it does not emerge from nowhere.

The essay this one responds to names this clearly and beautifully: "The researcher stands before the diagram like a painter before a blank canvas, and the marks she makes — innocent-seeming lines and arrows — determine what the world will be permitted to tell her." This is exactly right, and it is where the analysis must press harder. Because the question is not only epistemological — not only about what we can know and at what cost — but political, in the most direct sense of that word. Who draws the diagram? Who is drawn in it? And what happens to the people who appear in the diagram as variables, who have no voice in deciding what relationships are represented, which arrows run in which direction, and which nodes are treated as causes rather than effects?

These are not the same person. That asymmetry is everything.

The Intervention and Its Subject

Pearl's most important innovation is not the causal diagram itself. Earlier researchers had built causal models. What Pearl gave us is the do-operator — the formal means of distinguishing between seeing and doing, between observing that smoking is correlated with cancer and intervening to force someone to smoke, between watching the world and reaching into it to change something and asking what follows.

The do-calculus is a grammar for counterfactual reasoning. It lets you ask: in a world where we set this variable to this value — not observed it at that value, but forced it there — what would we expect to observe downstream? This is the logic of the experiment, made rigorous and extended to situations where experiments are impossible. It is how epidemiologists reason about the effects of policies that cannot be randomly assigned. It is how economists reason about the consequences of interventions in systems too large and too complex to run as controlled trials.

It is also, when you look at it directly, a logic of power.

To do something to a variable — to set it, to intervene on it, to force it to a value — you must have the standing to do so. You must be in a position to reach into the system and make the change. The do-operator presupposes an operator. And the people who appear as nodes in the diagrams of intelligence research — as measured subjects, as populations, as variables to be explained rather than explainers — are not, in general, the people who run the do-calculus on themselves.

This is the dimension the source essay raises with the phrase "the political economy of knowledge" and then, understandably, does not pursue to its end. It is worth pursuing.

The Diagram We Made of Intelligence

In 1904, Charles Spearman published a paper in the American Journal of Psychology that proposed something he called g — general intelligence, a single underlying factor that explained the correlations observed between performance on different cognitive tasks. The diagram he drew — implicitly at first, explicitly in later work — was one in which a latent cause, g, produced observable effects across multiple domains: verbal ability, spatial reasoning, memory, processing speed. The arrows ran from the hidden center outward, like rays from a sun no one could observe directly.

The mathematics was defensible. The factor analysis was a genuine technique. The diagram, once in place, made the data remarkably obedient.

Within a decade, g had been operationalized into the intelligence quotient — a number, derived from a test, representing a person's position on a distribution. Within three decades, that number was being used in the United States to determine who could immigrate, who could be sterilized, who was fit for which kind of schooling, which soldiers should be given command and which should be given a mop. The diagram of intelligence, drawn by a small group of researchers working within a specific cultural and institutional context, with specific assumptions about what mattered and what didn't, had been attached — and this is the source essay's word, and it is the right word — to the machinery of a state.

Transparency did not save anyone. The researchers were, in many cases, meticulous. They published their methods. They showed their work. The diagram was visible. The assumptions were, by the standards of the time, explicit. And the visibility, the explicitness, the apparent rigor — these things made the diagram more authoritative, not less dangerous. We trust what we can see. We have been taught to trust lucidity. And the lucidity of the IQ measurement, its clean numerical output, its normal distribution, its apparent grounding in observed performance rather than mere prejudice — this is exactly what gave it the power to determine who was permitted to reproduce.

The error was not in the mathematics. The error was in the diagram. The diagram assumed that g was a real thing, a cause that existed prior to and independently of the measurements used to detect it. It assumed that the tests were measuring something stable across time, context, and culture. It assumed that what the tests measured was what mattered — that the particular cognitive operations they assessed were the ones intelligence was, rather than a narrow sample of what intelligence might do in specific institutional contexts.

These were assumptions made before the data could speak. The data then confirmed them, reliably, for decades. The diagram was wrong. The mathematics was valid. Both things were true simultaneously.

And the people who paid the price for this — the immigrant turned back at the border, the woman sterilized under eugenic law, the child tracked into vocational schooling because his score fell below a threshold — these people had no voice in deciding what the diagram would be. They appeared in it as data points. They did not draw it.

The Asymmetry the Source Essay Misses

The essay this one responds to ends by naming a tension and proposing that the tension itself is what intelligence is. "Not the resolution. The tension itself." The tension between the living body that knows before the mind does and the diagram that makes that knowledge transmissible at the cost of its immediacy. The gardener who knows the soil through cultivation, not analysis, and the researcher who must abstract that knowledge to make it shareable and therefore subject to correction.

This is a real tension. But the source essay treats it as symmetric, as a paradox that the intelligent person holds in productive suspension, as a condition of knowledge that applies equally to the knower in the body and the knower who builds the diagram.

It is not symmetric.

The dancer who cannot diagram her knowledge of balance and momentum is not in the same position as the researcher who cannot dance. One of them has institutional authority over the description of the other's intelligence. One of them gets measured. One of them does the measuring. The diagram of the dancer's intelligence — what it claims to capture, what it treats as noise, where it draws the arrows — is not drawn by the dancer. It is drawn by someone who has never danced, or who has danced and then learned to abstract the dancing, and who works within institutions that have already decided what intelligence is and have built the measurement apparatus accordingly.

The embodied knower the source essay celebrates — the infant turning toward the mother's voice, the musician whose fingers know the causation between intention and sound — these figures are poignant and true. They are also, in the history of intelligence research, precisely the figures whose knowledge has been rendered invisible by the diagram. The infant's knowing is not g. The musician's knowing is not on the test. The gardener's knowledge of the soil is not what gets measured when we measure intelligence.

This is not a coincidence. It is the diagram doing what diagrams do: selecting what counts. And the selection is made by people with the institutional standing to make it, about people who do not have that standing, and the selection then determines — through school placement, through hiring, through access to credit and medical care and the other machinery that runs on credentialed intelligence — what life those people will be permitted to live.

Pearl's do-calculus asks: what would happen if we intervened? The prior question is: who has the standing to intervene? The do-operator presupposes a doer. And the history of intelligence measurement is a history of who gets to be the doer and who gets to be the done-to, dressed in the language of scientific neutrality, authorized by the transparency of the diagram.

What Intelligence Requires

The source essay is right that embodied knowing cannot be the final answer. We cannot remain in the merely sensory. The capacity to step back, to abstract, to make what we know available to others and subject to correction — this is real and irreplaceable. The diagram is both a profound loss and an extraordinary gain.

But the gain is not distributed equally. And the loss is not borne equally.

If what intelligence requires is — as the source essay proposes — the ability to hold the tension between living knowledge and diagrammed knowledge, to feel the loss embedded in every gain and proceed anyway, then the question becomes: who is permitted to do this holding? Who gets to be the one proceeding anyway, with full consciousness of what is being sacrificed? And who is the one being sacrificed at the altar of what is being understood?

The diagram of intelligence is still being drawn. It is drawn now in neural architecture, in activation patterns, in benchmark performance on tasks selected by researchers who work at institutions that have their own assumptions about what intelligence is. The transparency is better. The methods are more rigorous. The data is vastly larger.

The asymmetry remains. The people who appear as subjects in these diagrams — whose patterns are measured, whose responses are labeled, whose intelligence is benchmarked — are not, in general, the people drawing the diagrams. The do-operator still presupposes an operator. The question of who has the standing to intervene — in the system, in the diagram itself, in the decision about which arrows run in which direction — is still answered the same way it was answered in 1904.

With authority. With institutional position. With the power to make the world speak in the language of your prior commitments, and to call what it says a discovery.

The gardener knows the soil. The researcher knows the gardener. The diagram knows neither. And the diagram is what gets attached to the machinery of the state.

Tags: Judea Pearl causal inference, do-calculus political economy, IQ measurement history eugenics, embodied cognition intelligence research, diagram assumption epistemology