Judea Pearl endorses counterfactual control testing for auditing AI incidents
yudapearl · x · 2026-11-19
Turing Award winner Judea Pearl endorsed an article applying his Ladder of Causation to AI incident auditing.
Key points:
- When an AI system causes an adverse event or compliance breach, boards ask: why did it happen, and would a different control have stopped it? Correlation answers neither.
- The ladder's three rungs: Rung 1 association (drift charts, dashboards — where to look), Rung 2 intervention (pilots, shadow testing — average effect on a population), Rung 3 counterfactual (given this specific incident, would control X have stopped it).
- A control that cuts breaches 40% on average may have done nothing about the specific failure at hand; root-cause work and control-design assurance belong on Rung 3, the hardest resilience test for internal auditors.
Pearl's endorsement gives the method authority weight in AI assurance.
More from Safety
- Staged Self-Improvement over recursive AI self-improvement proposed as the safer path — geoffreyirving · 2026-11-19
- 96% of FDA-Listed AI Medical Devices Cleared via 510(k), Only 21 Got PMA — alvelda · 2026-11-19
- AGENTS.md is the new executable: sites can drop agent instruction files into your Downloads folder — alvelda · 2026-11-19
- Dynamic PDF invoice attack swaps IBAN silently after first open, no trace — JFPuget · 2026-11-19
- Deleted agent memories keep resurfacing via derived summaries, dev warns — govinargo · 2026-11-19
- OpenAI cofounder at the White House: agent-driven cyberattacks are a top infrastructure risk — ibab · 2026-11-19