AI will remove fake news and insults on Facebook


AI will remove fake news and insults on Facebook. It can cover from a thousand to a million publications.

Especially to remove insults and false news, researchers introduced SimSearchNet. This is a convolutional neural network that finds almost identical images or texts, therefore it automates the checks for which live moderators are usually responsible. For example, as soon as a social network employee notices that the post contains false claims about the coronavirus, this information will be sent to SimSearchNet – the AI ​​will look for duplicates or similar images. This method will increase the reach of moderators who can remove a thousand or even millions of posts in one go.

According to a Facebook blog entry, in preliminary tests, AI learned to correctly detect 88.8% of publications that “call for hatred.” Thus, the moderators of the social network did not have to manually delete about 9.6 million posts.

Researchers note that AI has just begun to learn. Some publications may contain references and nuances that can only be determined by a person or a non-standard-thinking device. In addition, the network will have to consider who writes the publication, to whom it is addressed, and what other content is attached to the post – for example, an image, audio, or video.

In addition, Facebook will continue experiments to improve natural language processing. The AI ​​will be able to translate the text into about 100 spoken languages ​​and determine which publications are supposed to violate the rules of the site.