It started with celebrity faces in adult content. The technology was disturbing but containable in a way that political deepfakes are not, because celebrity identity doesn't determine the future of governments. Political candidates do. And the same technology that put celebrity faces into fabricated videos has been applied to political figures, and the quality gap between detectable forgery and undetectable footage has essentially closed.
Deepfake detection technology is not keeping pace with deepfake generation technology. This is not a close race. Generation technology — the AI systems that produce synthetic video and audio — is improving with every hardware generation and every model release. Detection technology — the forensic tools designed to identify AI-generated media — is fighting a definitional problem: as generation improves, the artifacts that detection relies on become rarer and subtler until they effectively disappear.
In the 2024 election cycle, AI-generated audio of political figures was already deployed in robocall campaigns, in social media content, and in targeted disinformation operations against specific candidates in multiple countries. This is not speculation. These incidents were documented by election security researchers, reported by multiple news organizations, and in some cases led to regulatory investigations.
What has not been solved — what cannot be fully solved with current technology — is the attribution problem. When a convincing synthetic video of a political figure appears online, identifying its AI origin requires forensic analysis that takes days. The video spreads in hours. The retraction, when it comes, reaches a fraction of the original audience.
The asymmetry is the point. You don't need the forgery to be believed permanently. You only need the doubt to persist. Once someone has seen a sufficiently convincing synthetic video of a candidate saying something damaging, the denial never fully erases the memory. The neuroscience of this is well documented.
The deepfake election is already underway. The question is whether the audience knows how to watch.
SOURCE LINK: https://www.technologyreview.com/2024/01/22/1086433/deepfake-election-ai-generated-robocalls/
Comments
Post a Comment