Images and video generate more engagement on social networks but visual content is also harder to fact check. With machine learning advancements, however, Facebook expanded its fact-checking program from articles to photos and videos as well in an update that began rolling out on Thursday, September 13. That means the company is attempting to crack down on those fake political memes or images suggesting that yes, Hurricane Florence is creating a real-life Sharknado.
Artificial intelligence can already flag potential fake articles by looking for specific words and phrases, but it’s easier for a computer to read text than to “see” an image. Factoring in the fact that images can be “fake news” in more ways than one, finding fake photos and videos is an even tougher task.
Now, true to an earlier announcement, Facebook is expanding the fact-check program that already exists for articles to photos and videos. Advancements in machine learning will help flag potential fakes for review by a human fact-checker. Optical character recognition, for example, can recognize text saved in a JPEG instead of a text format. The program still relies heavily on human interaction, however, including user flags and key phrases within the comments.
Facebook divides fake images and video into three different categories. The first is the type most commonly associated with a fake photo — an image manipulated by software. As an example,
Doctored images aren’t the only fakes Facebook is looking for, however. Images can also be taken out of context — such as a photo of one event being shared with text that suggests it’s from an entirely different incident. The third type is unaltered photos with false accompanying text, or videos with false information in the audio.
Once a photo or video is flagged for review, the fact-checking organizations use clues like the metadata and conducting a reverse image search to try to find the original image. The organizations also use processes similar to fact-checking articles, including finding official reports. Like with links, Facebook also works to find the same fake presented from different sources and in multiple ways. With the expanded program, the network is also looking for the same misinformation crossing over content types.
The program is now expanding to 17 different countries, using Facebook’s 27 third-party fact-checking organizations. Facebook points out that different countries tend to have a different type of content that’s more widely shared than others. In some areas, articles are most shared, while others have more photos or videos.
Facebook pushes proven false articles lower in the news feed algorithms, removing the spread without censoring them entirely — so don’t stop looking for the signs a photo might be doctored.