# On Causation: Intelligence and the Architecture of Knowing ## The Banishment and Return Science, in its great recoil from speculation, stripped causal language from its instruments as one removes poison from a cup. For a century it called this amputation *rigor*. Correlation was permitted. Mechanism was not. The mind learned to speak only of associations, distributions, covariances—the shadows on the wall, never the fire that cast them. This was cowardice dressed as precision. Pearl restored what fear had exiled. He showed that causation is not mystical whisper but *structure*—a diagram made explicit, a set of assumptions about how the world is articulated. His tools are provably correct. But correctness is a narrow thing. It means: *given your diagram, these inferences follow*. It does not mean the diagram is true. Transparency here is a trap. A researcher may lay bare every assumption, every arrow, every conditional independence, and still be elaborately, completely, demonstrably wrong. ## The Diagram as Hidden Sovereign This is the concealed tyranny of causal reasoning: it requires you to *assume the structure before the data can speak*. The data does not reveal its own causation. The data obeys whatever architecture you impose upon it. You are not discovering the diagram. You are *installing* it. Consider: two researchers, both transparent, both rigorous, may draw different causal graphs of the same phenomenon. One places X as cause, Y as effect. Another reverses them. The mathematics works equally well for both. The data cannot adjudicate. Only *prior knowledge*—itself a form of assumption—can choose between them. And prior knowledge is precisely what is scarce when we study complex systems: intelligence, cognition, social behavior, disease. The diagram becomes a kind of invisible legislator. It determines what questions can be asked. It decides which variables are confounders, which are mediators, which are irrelevant. It permits certain causal claims and forbids others. Yet the diagram itself remains *assumed*, not proven. ## What It Means to Know a Cause To know the cause of something is to possess a *correct model of intervention*. Not mere prediction. Not mere association. But the ability to say: if I *do* this, *that* will follow. This is knowledge of a different order than pattern-matching. It is knowledge of *what happens when the world is altered*. Yet this knowledge has a peculiar fragility. It is local. It is conditional on the diagram being right. It is hostage to the variables you included and the variables you omitted. The cause you identify may be real within your model and false in the world. The cause may be true for the population you studied and false for another. The cause may operate through mechanisms you did not measure and therefore do not understand. Intelligence research illustrates this perfectly. Researchers have identified "causes" of intelligence: genetics, environment, education, nutrition, motivation. Each claim comes with a diagram. Each diagram assumes certain causal structures. The genetics-first diagram places genes upstream, environment downstream. The environmentalist diagram reverses priorities. The interactionist diagram tangles them together. The data does not choose. The diagram chooses, and the data obeys. ## Authority and the Question of Who Decides Here lies the political question hidden inside the technical one: *who decides what the diagram is*? In principle, the scientific community decides through debate and evidence. In practice, the diagram is decided by: - **Disciplinary convention.** Economists draw certain diagrams; sociologists draw others. The diagrams reflect the field's history, not nature's structure. - **Funding and incentive.** Research that requires expensive intervention gets different diagrams than research that merely observes. The diagram shapes what is fundable. - **Theoretical fashion.** At certain moments, certain causal structures are *intelligible* to a field and others are not. The diagram is partly an artifact of what the field has learned to think. - **Power and interest.** Who benefits from believing X causes Y rather than Y causes X? The diagram is never innocent. It has consequences. It justifies certain interventions and delegitimizes others. It empowers certain actors and constrains others. In intelligence research specifically: does intelligence cause achievement, or does achievement cause the measurement of intelligence? Does poverty cause low IQ, or does low IQ cause poverty? Does education cause intelligence, or does intelligence determine who receives education? These are not merely technical questions. They are *political* questions wearing technical disguises. The diagram you choose determines which interventions seem rational, which seem futile, which seem just, which seem cruel. ## The Dimension of Causal Pearl's great contribution was to make causation *explicit and mathematical*. This is genuine progress. But it is progress in the art of *stating assumptions*, not in the art of *validating them*. The causal dimension—the axis along which we arrange variables as causes and effects—is not discovered. It is *imposed*. It is a choice about how to carve up the world. Different carvings are possible. Each carving is internally consistent. Each can be stated with perfect clarity. But clarity is not truth. The researcher who is explicit about their causal assumptions is not thereby more rigorous than one who leaves assumptions implicit. They are only more *honest about their speculation*. Honesty is a virtue. It is not the same as correctness. ## Conclusion: The Limits of Causal Knowledge To know a cause is to possess a model that predicts intervention. To know that you know requires validating the model. To validate the model requires either: (a) experimentation—intervening in the world and observing consequences; or (b) comparison with other validated models. But in the study of intelligence, as in much of human science, true experimentation is often impossible, unethical, or prohibitively expensive. We are left with observation, comparison, and assumption. Under these constraints, causal knowledge is provisional, local, and contingent on diagrams we cannot fully validate. This is not a reason to abandon causal reasoning. Causal reasoning is indispensable. But it is a reason to remain *suspicious* of causal claims, especially when they are stated with certainty. The diagram is not the world. The diagram is a *tool for thinking about* the world. It is useful precisely because it is simplified, explicit, and manipulable. But its usefulness does not make it true. And its transparency does not make it valid. The researcher who forgets this distinction mistakes the map for the territory. The field that forgets it mistakes methodology for wisdom. Causation has been restored to science. But it has been restored as assumption, not as discovery. The burden of knowing what the diagram should be remains where it has always been: not with the data, but with the mind that interrogates it.