# On Cause, or What We Think We Know When We Think We Know Why There is a peculiar moment in reading when one becomes aware not of the words themselves—which have fallen away like the frame around a painting—but of the *structure* of thought moving beneath them. One feels, suddenly, the architecture of the author's mind. This is what happens, I think, when we confront Judea Pearl's work: we become aware, perhaps for the first time, that we have been living in a certain kind of house, and now someone has switched on the lights to show us the walls. For nearly a century, science spoke as though it had made a vow of silence regarding causes. It would describe correlations with exquisite precision. It would map the territories of association, the weather systems of variables moving in concert. But to ask *why* one thing moved another—to point and say "this caused that"—was to violate some unspoken contract between the knower and the known. It was considered imprecise. Unseemly. The province of philosophers or, worse, of certainty. And there was, let us acknowledge, something clarifying in this exile. By refusing the language of causation, science protected itself from a certain kind of careless thinking. It built ramparts against our natural hunger to explain. We wanted stories; it gave us equations. We wanted meaning; it gave us mathematics. For those trained to resist the seduction of narrative, this was bracing. Rigorous. One could almost hear the self-satisfaction in the word. But exile, when it lasts long enough, becomes a kind of erasure. One forgets what was forbidden and why. The prohibition becomes the whole truth. And so for generations of researchers—particularly in the social sciences, where the temptation toward causal explanation is nearly irresistible—there developed a curious double life: in their published work, they spoke only of associations, correlations, statistical relationships. But in their minds, in their intuitions, in the very architecture of their designs, they reasoned about causes. They thought causally while speaking correlational. They inhabited a kind of stuttering. Pearl's restoration of causal language is not, as one might suppose, a return to the unguarded speculation of earlier centuries. It is something more subtle and more troubling. He has given us tools—diagrams, calculi, theorems that are "provably correct." One can point to them, as one might point to a mathematical proof, and say: *here*. Here is rigor. Here is certainty. And yet. The diagram precedes the data. This is Pearl's honest acknowledgment of something that was always true but could be hidden when we spoke only in whispers about causes. Before any number can speak, before any correlation can emerge from the fog of observation, *someone* must draw a picture. Someone must decide which variables matter and which do not. Someone must trace arrows: this pointing to that, this *influencing* that. The diagram is not discovered in the data. It is *assumed*. It is *believed*. It is, in a word, chosen. Now, there is a particular kind of intellectual honesty in making explicit what was implicit. One might even call it progress. It is better, surely, to state one's assumptions boldly than to smuggle them in the back door. Yet honesty about the source of an assumption is not the same as correctness about it. A researcher can be "completely explicit about a model that is completely wrong." Yes. This sentence contains a kind of tragic poetry. It suggests that transparency is not innocence. That clarity is not truth. That one might draw a perfect diagram of an impossible world. And this is where the matter becomes, for me, genuinely difficult—not technically difficult, but difficult in the way that certain questions are difficult because they touch on power, on whose vision of the world gets to be the picture that shapes all subsequent interpretation. --- What does it mean to know the cause of something? In the physical sciences, we might approach this with something like confidence. We understand—or we think we understand—the mechanisms by which one billiard ball strikes another. The diagram seems almost to draw itself from nature. But even here, if one looks closely, one sees the fingerprints of the observer. We have chosen to see the world in terms of discrete objects, of time moving in one direction, of collision as a meaningful category. These are not mistakes, perhaps, but they are choices. Yet consider the social world. Consider, say, the cause of poverty, or of achievement, or of belonging. Here the diagram becomes vastly more complex. One researcher draws it thus: family background → educational opportunity → economic outcome. Another draws it thus: systemic inequality → differential access to capital → constrained choice architecture. A third draws it thus: individual motivation → effort expended → rewards accrued. All three diagrams can be made explicit. All three can be rendered mathematically. All three can claim a kind of rigor. But they are *different diagrams*. And the data—that supposedly neutral arbiter—cannot speak until one of these pictures has been placed before it like a stencil. The data will, in a sense, confirm whatever diagram has been chosen, provided the researcher has done the mathematics correctly. This is not because the data are dishonest. It is because data do not speak at all. They must be made to speak. And speech requires a language. The diagram is the language. Who decides what the diagram is? This is where the social dimension emerges with particular force. In the laboratory, where we can approximate control, where variables can be isolated and held still, the question of who draws the diagram feels almost abstract. But in the social world—in the world where human beings actually live, where causes are entangled with history, with power, with the accumulated weight of what was and what was not allowed to be—the question becomes urgent and political. A poverty researcher from an elite institution draws a diagram. A community organizer from the neighborhood draws a different one. Both are explicit. Both can be formal. Both can be "provably correct" *given their respective assumptions*. Yet they will lead to different interventions, different policies, different futures. One diagram suggests that we should improve access to education; another suggests that we should redistribute wealth; another suggests that we should change individual aspirations. Each diagram makes certain causes visible and renders others invisible. Each diagram, in other words, is a kind of power. And this is what troubles me most profoundly about the restoration of causal language, even when it comes with such admirable transparency: that it can make us forget—just when we most need to remember—that the choice of what to make visible is never itself a neutral, technical matter. It is an ethical and political choice, and it ought to be recognized as such. The diagram must be assumed before the data can speak. Yes. But we must also ask: who assumes? Who gets to draw? Whose vision of the world becomes the stencil through which all subsequent reality is filtered? Transparency about the diagram is essential. But it is not sufficient. For what we need—and this is the difficult part, the part that cannot be formalized into a calculus—is a kind of intellectual humility about the diagram itself. A recognition that the picture one has drawn, however explicitly, however rigorously, however correctly proven within its own logic, is still and always a picture. It is still a choice. It still excludes. It still privileges certain ways of seeing over others. The causal diagram, in other words, is not the cause. It is a representation of someone's theory of the cause. And when we forget this distinction—when we mistake the map for the territory—we commit a kind of subtle violence. We make certain people's stories invisible. We render certain suffering unintelligible within the terms of our equations. We close off certain futures because our diagram did not include them. This is not an argument against rigor. It is an argument for rigor of a different kind: rigor in acknowledging what cannot be formalized, what cannot be diagrammed, what must remain open to the testimony of those whose lives are lived within and against the categories we have chosen. The social world is the world where human beings decide things. Where power operates. Where the choice of what to make visible is itself a form of power. In this world, transparency is not validity. Explicitness is not truth. And the most rigorous diagram in the world cannot substitute for the harder work of asking: *whose diagram is this, and what does it render invisible?*