Attacking machine learning with adversarial examples
OpenAI Blog
Adversarial examples are inputs deliberately designed to trick machine learning models into producing incorrect outputs, functioning as optical illusions for AI systems. Researchers have demonstrated these attacks work across multiple input types, including images, audio, and text modalities. The difficulty in defending against such attacks stems from the fundamental challenge of making machine learning systems robust to intentionally crafted deceptive inputs that humans would recognize correctly.
Why it matters
Adversarial examples are inputs to machine learning models that an attacker has intentionally designed to cause the model to make a mistake; they’re like optical illusions for machines. In this post we’ll show how adversarial examples work across different mediums, and will discuss why securing systems against them can be difficult.
