The prospect of an AI-produced video of a world leader saying or doing something inflammatory leading to an election upset or triggering violence looks slightly more remote.
How it works: A new digital forensics technique uses machine learning to analyze an individual’s style of speech or movement. The system learns to distinguish the head and face movements that characterize the real person, for example the way they smirk, or nod their head.
The result? It was at least 92% accurate at spotting several variations of deepfakes in experiments.
An arms race: Deepfakes are now ridiculously easy to produce, and politicians are especially concerned that they could be used to sow misinformation during next year’s presidential election. Some tools for catching deepfake videos have already been produced, but forgers have quickly adapted. This new technique will be no different, unfortunately. Read the full story here.
—Will Knight
No comments:
Post a Comment