Of course, there is a margin of error in identifying artificial intelligence’s (metaphorical) fingerprint. While impressive, the technology to detect these AI-generated images isn’t perfect. Not all images have the metadata or watermarks that Meta relies on to identify AI.
While Meta’s AI lab, FAIR, is working on some promising watermarking tech called Stable Signature, only some AI-generated content can be traced. Bad actors may also find ways to remove watermarks or metadata or use tools that don’t add these markers in the first place.
Additionally, while many companies have begun including identification signals in AI image generators, the same can’t be said for AI tools that create audio and video content. Until the industry improves this capability, Meta is adding a feature to allow users to disclose whether audio and video content is AI-generated. Still, AI video has improvedin realism at an astounding pace, so we can expect more buzz around identification tools …