Meta plans to add a tool to Facebook, Instagram, and Threads that can spot fake pictures made by other companies’ AI. Right now, it already labels fake images made by its own AI.
They hope this will push the industry to fight fake AI images. But some experts doubt how effective it will be.
Sir Nick Clegg, a top Meta executive, says they’re working on this tech, but it’s not perfect yet. They want others in the industry to follow suit.
However, Professor Soheil Feizi from the University of Maryland says this tool might not catch all fake images. People could tweak images to dodge detection. Also, it might mistakenly flag real images as fake.
Meta admits this tool won’t work for fake audio or video, which are big concerns too. Instead, they’ll rely on users to label their own posts. If they don’t, there could be consequences.
Meta also can’t check for fake text, like what AI language models generate. Sir Nick Clegg says it’s too late for that.
The Oversight Board, which checks Meta’s decisions, says Meta’s rules on fake media aren’t clear or strong enough. They criticized Meta’s handling of a video of President Biden.
Meta’s policy on manipulated media isn’t good enough for the future, Sir Nick admits. They’re trying to update it.
Since January, Meta has required political ads to say if they use altered images or videos.
Source : BBC
https://www.bbc.com/news/technology-68215619