YouTube videos secretly enhanced with AI

YouTube has falsified published videos without disclosing this or informing their creators or uploaders. For months, the Google subsidiary has been reworking third-party videos with generative artificial intelligence, resulting in small but noticeable changes, such as strange hairstyles, incorrect skin tones or distorted body parts.

The British Broadcasting Corporation (BBC) has drawn attention to this. It was only after users complained in online posts and YouTube videos that YouTube spokesperson Rene Ritchie admitted the manipulations last week: "We are conducting an experiment on selected YouTube short videos," he said in an online post. The AI is supposed to reduce blurring and image noise, "comparable to what a modern smartphone does when recording video".

However, the AI sometimes takes a strange direction. And uploaders apparently cannot forbid YouTube's AI interventions in their videos.

Uploaders must label AI

In the fall of 2023, YouTube announced the introduction of the obligation to disclose and label AI-generated content. "In future, creators will have to disclose if they have created manipulated or synthetic content that appears realistic", the YouTube blog stated at the time. "This includes the use of AI tools."

However, this could only have meant major manipulation: "This could be, for example, an AI-generated video that realistically depicts an event that never happened. Or content that shows a person saying or doing something that they didn't actually say or do." The disfigurement of hair or make-up is a smaller caliber; for the person depicted, however, it can still be embarrassing or even damaging to business, for example if an author values authenticity.

On the other hand, the blog post explicitly referred to the "use of AI tools." The current regulation cannot be interpreted precisely. Heise online was unable to track down the exact specifications in the jungle of YouTube terms and conditions for authentic-looking AI products.

(ds)