Normal Accidents by Charles Perrow - Notes

Normal Accidents

📖 View Original Handwritten Notes (English Source)

Chapters 1 & 2 Chapter 3.1 Chapter 3.2

Download (Ch. 1-2)

Download (Ch. 3.1)

Download (Ch. 3.2)

Failure areas:

• Design
• Equipment
• Procedures
• Operators
• Supplies and material
• Environment

Failure origins:

1. Tightly coupled systems
2. Complexly interactive systems

“Seeing is not necessarily believing;
Sometimes, we must believe before we can see”

System (2 sides)

Reliability Graph

• Event A: Collapse of a medieval Cathedral
  High learning (at a time) → applied
• Event B: Nuclear plant accident
  Low learning (comparative) → hard to apply
• AI: perhaps we can use AI for better supervision & control on mission-critical objects leading to breaking plateau?

Operating experience

• Actual expansion / scale rate: 3.5%! → actually scaling 60x more aggressively (or) having 60x less experience with new tech

Process adherence issues:

• cover-ups
• intimidations
• falsifications
• neglect
• mistakes
• deception
• mistraining

Defining Accidents & Incidents

Incident vs. Accident?

• Definition: An accident is failure in system or subsystem, that damages more than one unit* and such disrupts the ongoing or future output of the system. * Argumentative. If system is disrupted, scale of root cause is (IMO) irrelevant. Second sentence prevails over first.
• Relativity: Same event can be considered an accident in one system while not in a supersystem (abstraction / scale layers).
• Disrupts: Requires direct immediate action.
• Human(s): Not alternative class, just a subsystem. Therefore damage to humans (Level 3) = accident.

Questioning Defense in Depth?

Victims:

Level of involvement into system (Highest to lowest)

• Risk and Pay: No/inverse correlation between risk & pay across industries.
• System Users vs. Bystanders (implicit): First to some extent “choose” to participate => elect to share (some) risk. But users, unlike operators, have no influence over system operation.
• Fourth-Party Victims: Continuous neglect over fourth party victims, now vs future.

Interactions: Linear / Complex

• Engineered Safety Devices (ESD): Can’t protect from non-obvious interactions.
• Monitoring & Corrective Actions: Hidden interactions, monitoring means (too few vs too many) => potential for “wrong” corrective action leading to more failures.