Estimating the Brittleness of AI: Safety Integrity Levels and the Need for Testing Out-Of-Distribution Performance

TL;DR

This paper highlights the challenges in evaluating AI safety, emphasizing the difficulty in estimating system brittleness, especially outside training distributions, and calls for improved testing and resilience strategies.

Contribution

It introduces the concept of brittleness in AI systems, demonstrating the limitations of current performance metrics and stressing the importance of OOD testing for safety certification.

Findings

AI systems are more failure-prone than traditional critical systems within known bounds.

Performance degrades gradually as inputs become Out-Of-Distribution.

Enhanced resilience and OOD evaluation are crucial for AI safety certification.

Abstract

Test, Evaluation, Verification, and Validation (TEVV) for Artificial Intelligence (AI) is a challenge that threatens to limit the economic and societal rewards that AI researchers have devoted themselves to producing. A central task of TEVV for AI is estimating brittleness, where brittleness implies that the system functions well within some bounds and poorly outside of those bounds. This paper argues that neither of those criteria are certain of Deep Neural Networks. First, highly touted AI successes (eg. image classification and speech recognition) are orders of magnitude more failure-prone than are typically certified in critical systems even within design bounds (perfectly in-distribution sampling). Second, performance falls off only gradually as inputs become further Out-Of-Distribution (OOD). Enhanced emphasis is needed on designing systems that are resilient despite failure-prone AI components as well as on evaluating and improving OOD performance in order to get AI to where it can clear the challenging hurdles of TEVV and certification.