Fake news often lives inside Facebook through photographs and videos created on purpose to spread disinformation campaigns, rather than being carried outside the website in news articles that could generate advertising revenue. To fight these intermediaries driven more by political questions than by financial motives, the social network decided it really must be able to detect fake content carried by images and even by the audio that accompanies video clips.
Rita Hayworth was a zombie and Hillary Clinton adopted an alien in these fake news stories
This week, the tech giant announced that it is expanding its fake-news detection system through a programme that was already being used in four countries. Now 23 nations will have access to this method. "Many of our external partners who do this kind of detection have extensive experience in evaluating photos and videos and are trained in visual verification techniques, such as reverse image search and metadata analysis, which tell us, for example, when and where the photo or video was taken", said Facebook product manager Antonia Woodford. "The more evaluations we have from our analysts, the more we will be able to improve the accuracy of our artificial intelligence machine that learns that process".
Fake news detected by analysts of manipulated images
The main goal of the world's largest social network is to be able to automatically detect images that have been manipulated, taken out of context and that do not show what they claim to show, or text and sound that are very likely fake. In a 3,260-word manifesto on security, Mark Zuckerberg explained this week that the "definition of success is being able to stop cyberattacks and coordinated information operations before they cause any harm". That means using artificial intelligence to chase fake news, instead of waiting for it to be reported by users. To do that, Facebook needs to train the machine that collects the data to exhaustion, through the human partners who take part in these detection operations.
Fake posts circulated through Facebook
The American platform is therefore developing technological tools to assist professionals in this field during the process. "We are using optical character recognition to extract text from photographs and compare it with the article headlines from the fact-checkers. We are also working on new ways to detect whether the photo or video has been manipulated", says Woodford, referring to the Reddit DeepFakes user, who uses video-editing software to make someone appear to be saying or doing something they were not, in reality, doing. Memes were one of the most popular forms of disinformation used to interfere in globally relevant issues.
The problem is that they are so easily shareable that people do not even need to leave Facebook to see them, and they reach viral distribution from unsuspecting users who do not even realise they have become pawns in manipulation campaigns. Zuckerberg's network can potentially use resources that demand a high technological level to build detectors of fake news shared through memes, giving an argument for keeping the platform running. Combining Facebook, Messenger, Instagram and WhatsApp, the company gains economies of scale in the fight against what is already considered a plague of false information.
