Is AI Robust Enough for Scientific Research?

TL;DR

This paper reveals that neural networks used in scientific research are highly vulnerable to small perturbations, posing risks to their reliability across diverse scientific applications.

Contribution

It uncovers a broad, general vulnerability of neural networks to minute perturbations in scientific applications, highlighting a critical reliability concern.

Findings

Neural networks are highly susceptible to small input perturbations.

Vulnerability observed across five diverse scientific domains.

Raises concerns about the reliability of AI in scientific research.

Abstract

We uncover a phenomenon largely overlooked by the scientific community utilizing AI: neural networks exhibit high susceptibility to minute perturbations, resulting in significant deviations in their outputs. Through an analysis of five diverse application areas -- weather forecasting, chemical energy and force calculations, fluid dynamics, quantum chromodynamics, and wireless communication -- we demonstrate that this vulnerability is a broad and general characteristic of AI systems. This revelation exposes a hidden risk in relying on neural networks for essential scientific computations, calling further studies on their reliability and security.