Threshold Crossings as Tail Events for Catastrophic AI Risk

Elija Perrier · 2025

Read Threshold Crossings as Tail Events for Catastrophic AI Risk by Elija Perrier (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-18979.txt.

Description

Elija Perrier analyzes how threshold crossings in AI systems can act as tail events for catastrophic risk. The 2025 paper studies bifurcation-driven jumps, where a small change pushes a system into a very different state.

When a control parameter fluctuates randomly near a catastrophic threshold, the outcomes can follow a heavy-tailed distribution. Perrier shows when the chance of a sudden, large-scale transition aligns closely with the tail probability of the damage distribution. That link matters for how risk is estimated. If damage tails and transition chances match, then monitoring the tail gives a direct handle on the chance of the jump itself. The work aids monitoring and control of such systems.

Catalog record

CREATORElija Perrier
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE16 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-18979

Browse the full library · About the Tower of Babel