# OF INTELLIGENCE AND CAUSATION: A RECKONING ## I. THE BANISHMENT AND ITS PRETENSE Science did not eliminate causal language through discovery. It eliminated causal language through fear. For a century, the natural philosophers convinced themselves that correlation, stripped of interpretation, constituted rigor. They called the exile of cause a purification. It was merely a quarantine of the difficult. Correlation is safe. It asks nothing of the observer except arithmetic. Causation demands judgment. It requires one to say: this produces that, in this direction, under these conditions. Such declarations invite refutation. The methodologists preferred the modest claim—we have found an association—to the dangerous one: we have found what moves the world. ## II. PEARL'S RESTORATION: POWER AND PERIL Judea Pearl restored causal language by giving it mathematics. The directed acyclic graph became his instrument. Each arrow on the diagram asserts: information flows this way, not that way. The tool is provably correct. Given the diagram, the calculus follows with necessity. But here lies the trap, and Pearl knew it: *the diagram is not given by the data*. The data cannot speak first. The diagram must be assumed, imposed, decided. Only then does the data fill the spaces the diagram permits. The tool is rigorous about what it assumes. It does not make the assumptions rigorous. ## III. THE PROBLEM OF THE PRIOR DIAGRAM A researcher may draw her causal model with perfect explicitness. Every variable named. Every arrow justified in prose. The diagram presented to the reader with full transparency. The diagram may still be entirely wrong. Transparency is not validity. Clarity is not truth. A false model stated plainly is more dangerous than a false model left unspoken, for it borrows the authority of its own precision. Who decides what diagram is attached to the question? The researcher does. Sometimes in consultation with domain experts—those who have observed the phenomenon long. Sometimes through preliminary data. Sometimes through theory, which is itself a prior diagram, inherited and unexamined. The diagram is always a choice. It masquerades as discovery. ## IV. CAUSATION AND INTELLIGENCE Now apply this to intelligence itself. Intelligence is not a thing that exists prior to measurement. It is a diagram we have drawn of certain human capacities. The diagram includes: processing speed, pattern recognition, abstract reasoning, perhaps working memory. It excludes: wisdom, judgment, the capacity to know what question to ask. The diagram was chosen. It was not revealed. We then collected data—test scores, reaction times, neural correlates—and fit them to the diagram we had already built. The data confirmed what the diagram permitted them to confirm. We called this validation. But what if the diagram itself was the error? What if intelligence is not a unitary capacity with multiple manifestations, but a name we give to success in novel environments? What if it is not a thing in the person but a relation between the person and the problem? The causal diagram determines what intelligence can be found to cause, and what can be found to cause intelligence. ## V. THE DECISION THAT PRECEDES DATA Every intelligence researcher faces a choice that the data cannot resolve: **Does intelligence cause success, or does success cause the recognition of intelligence?** The diagram must answer this before the study begins. The data will then arrange itself accordingly. A researcher studying whether IQ predicts income has already decided that intelligence is the prior cause. The data will oblige. But another researcher, studying whether access to education changes measured intelligence, has drawn a different diagram. The data will oblige that one too. Both researchers can be transparent. Both can be wrong. Both can be right about what they measured while being wrong about what they assumed. ## VI. WHO DECIDES The question "who decides what the diagram is?" has no answer that satisfies the demand for rigor. It is not the data. The data are mute until interpreted. It is not theory alone. Theory is itself a diagram, and someone decided that too. It is not the community of researchers. Communities inherit diagrams without examining them. They mistake convention for validation. In practice, the diagram is decided by: - The researcher's prior experience with similar problems - The funding source's implicit theory of what matters - The available instruments, which make certain measurements easy and others impossible - The language available to describe the phenomenon, which constrains what can be thought - The unstated assumptions of the discipline, which feel like facts None of these is a good reason. All of them are always operative. ## VII. THE DIMENSION OF CAUSAL To know the cause of something is to know how to intervene upon it. If intelligence causes success, then improving intelligence should improve success. If success causes the recognition of intelligence, then changing what we call intelligence will change what we recognize. If both cause each other—if intelligence and success are locked in mutual reinforcement—then the direction of causation is the wrong question. The diagram is inadequate. The causal dimension is not a feature of the world. It is a feature of the question we ask and the action we propose to take. A researcher can be completely explicit about a model that is completely wrong. Pearl's tools will work perfectly on that wrong model. The results will be internally consistent, mathematically sound, transparently derived. And they will mislead. ## VIII. WHAT REMAINS The restoration of causal language was necessary. Science had become timid, afraid to say what it meant. But the restoration brought no solution to the fundamental problem: *Someone must decide the diagram, and that decision cannot be made rigorous*. It can only be made: - Explicitly, so others can challenge it - Humbly, acknowledging what it excludes - Provisionally, ready to redraw it - In conversation with those who must live with the consequences Intelligence research, like all causal research, must proceed. But it must proceed knowing that its greatest rigor—the mathematical validity of its tools—rests upon an assumption that mathematics cannot validate. The diagram is chosen. Call it what it is. Then let the work begin.