Adversarial examples are a term used in the fields of artificial intelligence, cybercrime, cybersecurity and digital transformation. They describe a special type of data that has been deliberately altered in such a way that it intentionally misleads machines such as artificial intelligence (AI).
Imagine a camera automatically checks whether a stop sign is visible in a photo. A so-called adversarial example would be an image that clearly shows a stop sign to our human eye. However, if the image is deliberately manipulated by small, usually invisible changes, the AI can suddenly no longer recognise that it is a stop sign - even though a human would not notice the difference.
This phenomenon is important for companies to be aware of because hackers can use adversarial examples to outwit facial recognition systems or other automated checks, for example. This is why the topics of security and protection against such attacks are particularly relevant when developing AI systems. When using artificial intelligence, it is worth thinking not only about the function, but also about protection against adversarial examples.
Deutsch 
English (UK)