Normal Accidents

Type: Cognitive Bias / Systems
Local HTML: normal_accidents.html

---

Definition

Complex systems fail inevitably — accidents are normal, not exceptional. Tight coupling + interactive complexity = inevitable failures.

Charles Perrow (1984): In complex, tightly-coupled systems, multiple small failures interact in unexpected ways. No one cause — systemic properties cause accidents.

---

Why It Matters

Nuclear plants: Three Mile Island — multiple small failures cascaded. Financial systems: 2008 crisis — interconnected derivatives amplified failures. Air travel: Rare crashes, but when systems fail, they fail catastrophically. Software: Microservices fail in unexpected interaction patterns.

---

Two Dimensions

	Loose Coupling	Tight Coupling
Complex	Recoverable (university)	Normal accidents (nuclear plant)
Linear	Manageable (assembly line)	Brittle (dams)

Tight coupling: Changes propagate fast, no time to recover Interactive complexity: Unexpected interactions between components

---

The Implication

You cannot prevent all accidents in complex, tightly-coupled systems.

You can:

• Reduce complexity
• Loosen coupling
• Contain failures
• Learn from near-misses

But accidents will still happen. They’re features of the system, not bugs.

---

Related Biases

• [[Black Swan** — Unpredictable catastrophic events
• [[Illusion of Control** — Thinking we can prevent all failures
• [[Optimism Bias** — “It won’t happen here”

---

Audio

Podcast episode: Normal Accidents

---

Part of the Cognitive Bias Reference