Cause
Hume noticed in 1739 that we have never, in the entire history of human perception, actually observed causation. We see the cue ball strike the eight ball. We see the eight ball move. We see this happen reliably, every time, for as long as we have watched billiards. What we never see is the connection between the two events. The connection is supplied by the mind, which has noticed the pattern often enough to expect the second event when the first occurs. Cause is a habit of the brain, projected outward onto a world that contains, strictly speaking, only conjunctions. The world doesn’t push. The mind interprets.
This was a wrecking ball, and it has not really stopped swinging since. Kant tried to rebuild the foundations by making causation a built-in category of the mind rather than a feature of the world. Russell, two centuries later, suggested that mature physics didn’t need the concept at all — there were only differential equations relating states, and cause was a primitive folk-physics word that working scientists had quietly outgrown. The folk word survived anyway, because we had nothing else with which to talk about anything.
Modern probabilistic causation, mostly Judea Pearl’s work, gave us tools that didn’t exist when Hume was writing. Do-calculus lets you reason rigorously about interventions: not just when X happens, Y tends to follow, but if I were to make X happen, would Y still follow? The distinction matters because most of what gets called “causal inference” in everyday science is actually correlation that has been politely renamed. Pearl’s work is the difference between people who carry lighters get lung cancer and carrying a lighter causes lung cancer. Both statements can be true in the data and only one of them survives the intervention test.
The law has its own fight with the same problem. The “but for” test asks: but for the defendant’s action, would the harm have occurred? It sounds simple. It is not. But for the killer’s bullet, the victim would still be alive, fine. But for the failed brake design, the crash would have killed someone else, slightly later, in a slightly different way — does the manufacturer’s negligence cause this death, or merely change which death occurred? The legal system has spent two thousand years building partial answers and has not fully resolved the underlying problem because the underlying problem cannot be fully resolved.
What everyday usage hides is how rarely we actually have the evidence to claim a cause. Most of the causes in conversation are stories — narratively satisfying connections between events that the speaker has noticed in temporal sequence. I got sick because I was stressed. The company failed because the CEO didn’t listen. The marriage ended because we wanted different things. These statements may be true. They may also be the narrative fallacy doing what it always does — supplying a single tidy thread to a tangle of partial causes that no one can fully see, because no one is conducting the experiments that would tell them apart.
This is why falsification mattered to Popper, and why the streetlight effect matters in research. We look for causes where the looking is cheap. The places where causes actually live are mostly dark and expensive to enter. Most of the world’s causal structure is invisible to us not because it isn’t there but because the experiments that would reveal it are unethical, impossible, or have never occurred to anyone. The library’s notes on map and territory and apophenia are partial defenses against a problem that has no full defense.
The instruction the concept eventually offers is uncomfortable. When you find yourself confidently asserting that X caused Y, ask: by what experiment could you have known? If the experiment is impossible — and it usually is — you have not stated a fact about the world. You have stated a story you find satisfying about the world. Sometimes the story is true. Often it is one of several stories that would have been equally satisfying given the same data. The honest position is to keep the story but flag it as one. Most of the universe is run by causes nobody can see, and the ones we see were almost all inferred under streetlights.