When a neutron star merger sends gravitational waves rippling across spacetime, LIGO detects them 1.3 billion years later. We say the merger caused the detection. But what does this claim actually assert? Nothing in the fundamental equations of general relativity mentions causation—only geometry, symmetry, and correlation. Yet we cannot describe the physical world, prosecute crimes, or design experiments without invoking causal relations at every turn.

This tension sits at the heart of contemporary metaphysics. Causation appears indispensable to scientific practice while remaining strangely absent from our best fundamental theories. Bertrand Russell famously argued that causation is a folk notion that mature physics should abandon, like phlogiston. Yet a century later, causal reasoning has proliferated rather than receded, becoming central to epidemiology, econometrics, and even machine learning through the interventionist frameworks of Pearl and Woodward.

The philosophical stakes are considerable. How we analyze causation shapes what we can say about scientific explanation, whether backward causation is coherent, how moral responsibility attaches to agents, and why time seems to flow in one direction. This article maps four dominant theories—Humean regularity, counterfactual dependence, causal processes, and interventionism—stress-tests them against notorious puzzle cases, and argues that the pluralism emerging from these debates is not a failure but a diagnostic feature of causation itself.

Hume's regularity theory launched the modern debate by identifying causation with constant conjunction plus temporal priority. C causes E just in case events of type C are regularly followed by events of type E. The account is admirably naturalistic—it demands nothing beyond what science already tracks—but it notoriously conflates causation with mere correlation, cannot distinguish causes from effects on any single occurrence, and leaves the modal force of causal claims mysterious.

David Lewis's counterfactual analysis, first developed in 1973, addressed these deficits by grounding causation in counterfactual dependence: C caused E if, had C not occurred, E would not have occurred. Cashed out through possible-worlds semantics, this framework elegantly captures singular causation and distinguishes causes from mere correlates. It also opens causation to formal treatment via similarity metrics across worlds, though it inherits controversy about how such metrics are fixed non-circularly.

Wesley Salmon and Phil Dowe pushed in a different, more physicalist direction. Their process theories identify causation with the transmission of conserved quantities—energy, momentum, charge—along continuous spatiotemporal worldlines. This grounds causation in physics itself: a genuine causal process transmits a mark, whereas a pseudo-process like a shadow does not. The theory shines for paradigm physical interactions but strains when applied to omissions, absences, and higher-level causal claims in biology or economics.

James Woodward's interventionism, developed alongside Judea Pearl's causal graphs, defines causation via hypothetical interventions: C causes E relative to a variable set if some intervention on C would change the distribution of E. This manipulationist framework has become dominant in the special sciences precisely because it aligns with experimental methodology. Yet critics worry it presupposes agency or renders fundamental physics acausal, since intervening on the initial conditions of the universe is not obviously coherent.

Each theory captures something real. Regularity tracks nomological structure, counterfactuals capture modal dependence, processes anchor causation in physics, and interventions link causation to inquiry. The question is whether these are competing analyses of one phenomenon or lenses onto genuinely distinct causal notions operating at different levels of reality.

Takeaway

When four sophisticated theories carve up the same phenomenon differently and each illuminates real cases, the pluralism may be metaphysically informative rather than a sign that three of them are wrong.

Preemption cases have been the graveyard of many causal theories. Suzy and Billy both throw rocks at a bottle; Suzy's strikes first and shatters it, while Billy's flies through empty air. Suzy caused the shattering. But had Suzy not thrown, Billy's rock would have shattered the bottle anyway—so counterfactual dependence fails. Lewis responded with elaborate machinery involving stepwise dependence and quasi-dependence, but each patch invites new counterexamples in an escalating dialectic.

Overdetermination compounds the problem. Two assassins fire simultaneously; both bullets strike the victim's heart. Each shot was sufficient, neither was necessary, and yet we naturally judge both to be causes. Regularity theories handle this poorly because the effect would have occurred without either regularity being instantiated singly. Process theories fare better here, tracing distinct energy-momentum transmissions from each rifle, though this suggests process accounts may be more fundamental for physical causation specifically.

Absence causation may be the sharpest test. The gardener's failure to water caused the plant's death. But an absence is not an event, transmits no conserved quantity, and stands in no spatiotemporal relations. Salmon-Dowe process theories simply cannot accommodate such cases, which is fatal given that biology, medicine, and law routinely traffic in omissions. Interventionism handles absences smoothly by treating them as values of variables, but at the cost of admitting non-physical relata into the causal fabric.

Double prevention introduces further complications. A pilot bombs a target because a wingman shoots down an enemy fighter that would have shot down the bomber. The wingman's action caused the bombing by preventing a prevention—yet no continuous process connects the two. Such cases are ubiquitous in biology, where inhibitors of inhibitors constitute much of cellular signaling, suggesting process purism is inadequate for the life sciences.

These puzzle cases are not marginal curiosities. They reveal that our concept of causation packs together dependence, production, transmission, and manipulability—features that come apart in principled ways. A theory that handles preemption may stumble on absences; one that captures omissions may lose grip on physical production. This is diagnostic evidence that causation is not one thing.

Takeaway

When counterexamples cluster systematically rather than randomly, they are not defects to be patched but signals that a concept has internal joints along which it can be legitimately carved.

Scientific explanation stands or falls on causation. The deductive-nomological model treated explanation as subsumption under laws, but Salmon and others showed that mere derivation from laws cannot distinguish explanation from mere prediction—the flagpole explains the shadow's length but not vice versa, despite their symmetric derivability. Only causal asymmetry breaks the tie. Contemporary work in causal modeling has vindicated this insight: to explain is to situate a phenomenon within a network of causal dependencies, not merely to derive it.

The metaphysics of free will and moral responsibility depends critically on how we analyze causation. Compatibilists like Frankfurt and Fischer argue that responsibility requires only the right kind of causal history—reasons-responsive processes flowing through the agent. Libertarians demand agent causation as a sui generis relation. Hard determinists insist that if all our actions are causally determined by prior states, responsibility evaporates. Which of these positions is defensible depends on whether causation is production, dependence, or manipulation.

The asymmetry of time may itself be causal. Physical laws are largely time-symmetric, yet causes precede effects with striking regularity. Some philosophers, following Reichenbach and Price, argue that the causal arrow is grounded in the thermodynamic arrow—that entropy gradients underwrite both the direction of time and the direction of causation. If correct, this would explain why causal reasoning works, why memory is asymmetric, and why we can influence the future but not the past.

Getting causation right also matters practically. Causal inference now drives epidemiology, policy evaluation, and machine learning through frameworks like Pearl's do-calculus. Distinguishing correlation from causation is not academic when trillions of dollars in medical decisions hang on the difference. The recent revolution in causal AI, which augments predictive models with structural causal models, is essentially applied metaphysics—a demonstration that abstract analyses of causation have surprisingly concrete payoffs.

Even fundamental physics is being reconsidered through causal lenses. Recent work on causal set theory attempts to reconstruct spacetime from primitive causal relations. Interpretations of quantum mechanics differ crucially on whether measurement outcomes cause or are merely correlated with distant results. The causation question is not settled decoration atop physics—it is increasingly load-bearing.

Takeaway

Metaphysical questions that appear abstract often turn out to be structural: how we answer them determines what we can build, prove, or justify in domains that seemed far removed.

Causation resists reduction to any single analysis because it does genuine work at multiple levels of reality. Process theories capture something true about physical production; counterfactual and interventionist accounts illuminate the dependencies that guide inquiry and action; regularity theories track the nomological patterns that make prediction possible. The proliferation of theories mirrors the proliferation of causal roles.

This pluralism need not collapse into anti-realism. A naturalistic metaphysics can hold that causation is a real feature of the world while acknowledging that different theoretical frameworks illuminate different aspects of it—much as wave and particle descriptions each capture genuine features of quantum systems without either being complete.

As physics probes the emergence of spacetime from more primitive structures, and as the special sciences develop increasingly sophisticated tools for causal inference, the metaphysics of causation becomes not less but more urgent. The question is not whether causation exists, but what kind of thing must exist for our best science and our deepest moral practices to make sense together.