From Gwern.net

Why Correlation Usually ≠ Causation

Everyone repeats that correlation isn't causation; almost nobody can say why the two come apart so reliably. Gwern's answer is structural — in a dense causal network the number of paths that merely correlate two variables explodes far faster than the narrow set that would let you intervene on one and move the other — and it reframes the replication crisis as arithmetic rather than misconduct.