Facebook’s AI Failure
How Meta's platform is failing to deal with AI-generated images of fake Americans in the US elections
New research by CCDH shows that convincing AI-generated images of political endorsements made by fake Americans are getting millions of interactions on Facebook ahead of the US elections.
About
AI-generated images of fake Americans appearing to make political endorsements, some of which portray fake military veterans, are gaining millions of likes, comments and shares on Facebook in the run up to the US presidential elections.
Comments on the posts show that many Facebook users appear to find the images credible. They include photorealistic images of fake military veterans, police officers and protestors appearing to endorse political candidates or policies.
Despite Facebook’s promise to label AI-generated images, none of the images identified by researchers were labelled. Transparency information shows that Pages posting the images are in some cases being run from Morrocco or Pakistan.
Researchers identified 169 posts published since July 1 with a combined total of 2.4 million interactions. These posts were shared 476,014 times. All posts contained an image of a fake person offering political opinions relevant to the US election and were assessed by researchers to be likely made using generative AI.