Probabilistic Risk Formalization

AI Risk Probability Decomposer

Claims of "0% chance" or "90% p(doom)" treat existential AI risk as a single monolithic gamble. Rigorous decision analysis decomposes the hazard into a causal conjunctive chain: frontier autonomy, goal misalignment, defense failure, real-world physical leverage, and multilateral containment breakdown.

Hypothesis Presets:

Probability Survival Funnel (Causal Pipeline)

Intervention Leverage (Tornado) Δ Risk from 50% Safeguard

Monte Carlo Distribution (10k Runs) 10,000 synthetic trials

0% Median: 4.6% 25%+

Key Causal Finding: The thesis that catastrophic risk is near-zero relies entirely on Physical & Cyber Leverage being gated by embodiment bottlenecks, or Societal Defense Dominance immediately neutralizing rogue agents.

Model parameters loaded and verified.

Analytical Framework: Why Single-Number Probabilities Mislead

1. The Conjunctive Rule

An existential catastrophe requires every single defensive layer to fail in sequence: P(Doom) = P(A) × P(B|A) × P(C|A,B) × P(D|A,B,C) × P(E|A,B,C,D). If any single barrier holds (such as robust air-gaps, hardware governance, or defensive AI superiority), overall risk collapses exponentially toward 0%.

2. Anatomy of the "0% Claim"

Computer scientists claiming a "0% chance" (e.g. Yann LeCun) do not necessarily deny AI capabilities. Instead, their model sets the conditional probability of uncontrollable self-directed drive or physical sovereignty to zero, positing that goal-driven architectures remain subordinate to hierarchical objective functions.

3. Where Interventions Matter

The sensitivity tornado identifies where research funding or governance produces the greatest variance. Improving post-deployment defense and monitoring often provides higher marginal leverage than arguing over whether AGI arrives in 2028 versus 2035.