EXECUTIVE SUMMARY:

Social media companies have been searching for better ways to identify, and remove harmful online content. Human moderation of social media content is not only time consuming and costly, but it also affects the mental health of the workers subjected to content moderation. One solution consists of creating new machine learning systems that can independently moderate content.

Facebook’s latest community standards enforcement report stated that these programs now allow for the removal of 98% of terrorist videos and photos before user exposure.

Facebook’s AI programs use two approaches to monitor for harmful content. First, they use what have been dubbed as “neural networks,” which can search for features and behavior of known objects and then label them. Second, Facebook uses a hash, or a unique string of numbers within a computer code to label the content as harmful. It then deletes any current copies throughout the system and prevents users from trying to re-upload the same content.

While this automated moderation program currently processes only 16% of posts violating Facebook’s community standards, it represents a promising means of better moderating content on social media networks.

To learn more about Facebook’s efforts to moderate harmful content through machine-learning systems, see this article by the MIT Technology Review.