What Is Causation?

You strike a match and it ignites. The striking caused the flame—this feels as obvious as anything can be. But David Hume, the eighteenth-century Scottish philosopher, noticed something unsettling: we never actually see causation. We see the striking. We see the flame. We see that the second follows the first, reliably, every time. But the “causing”—the invisible link between them—is something we infer, not something we observe. Hume concluded that causation might be nothing more than regularity plus temporal order plus our expectation that the pattern will continue. No metaphysical glue. Just sequence, habit, and projection.

Three Lenses

Regularity. On this view, “A causes B” means nothing more than: when A occurs, B reliably follows. Constant conjunction, as Hume put it. There is no extra “causal force” beyond the reliable sequence. This view is clean and minimal, but it struggles with the difference between correlation and causation—the rooster crows before sunrise every morning, but it does not cause the dawn.

Information flow. On this view, A causes B when knowing A reduces uncertainty about B—when A carries information about B that nothing else in the context provides. Judea Pearl, a computer scientist at UCLA, formalized this through causal graphs (directed acyclic graphs, or DAGs) and interventionist logic: A causes B if intervening on A changes B, holding everything else constant. In other words, causation is directed information transfer—constraint flowing from one state to the next.

Mechanism. On this view, causation is not just “what follows what” but “how A leads to B.” The mechanism is the process—the intermediate steps, the physical or logical chain connecting the two. Stuart Glennan, a philosopher at Butler University, and others have developed mechanistic accounts that ground causation in the actual structure of the process. We want more than regularity. We want the gears.

Convergence

These three views appear to disagree, but they converge on a deeper point. All three suggest that causation is not a primitive, irreducible force in the world—a metaphysical glue holding events together. It is, rather, our way of describing the fact that the state of the world at time t constrains the state of the world at time t+1. Physical laws plus initial conditions determine what happens next. “Cause” is the epistemic act of picking out which parts of that constraint matter for the explanation we are constructing.

When we say “the match strike caused the flame,” we are selecting, from an enormous web of conditions (oxygen present, match dry, surface rough enough, temperature below the ignition point of the surroundings), the one variable we care about for the purpose at hand. The cause is real—intervention on the strike changes the outcome. But the causal claim is shaped by our explanatory interests, not just by the physics.

This is provisional. Causation remains one of philosophy’s hardest problems. But the convergence across regularity, information, and mechanism frameworks toward an epistemic, constraint-based view is significant. It suggests that the question “what is causation?” may have been partly the wrong question—that the right question is “what are we doing when we identify a cause?”—and the answer is: picking out the features of constraint that matter for our purposes.

How This Was Decoded

This essay integrates David Hume’s regularity theory, Judea Pearl’s interventionist causal framework at UCLA, mechanistic philosophy of causation (Stuart Glennan, Peter Machamer), and information-theoretic approaches to causation. Cross-verified: all three frameworks converge on a constraint-based, epistemic characterization. The same structure appears in physics (laws as constraints on state evolution), biology (mechanisms as causal chains), and statistics (intervention as the test of causation).