Facebook Offers $50K For AI That Can Detect Hateful Memes
Facebook prides themselves in delivering content on their site that is pre-filtered for hate speech. However, the commonly shared meme is a multimodal content that Facebook’a AI struggles to distinguish the meaning of. Currently, Facebook is offering a 50,000 dollar prize to a winner who can develop an AI that detects hateful speech in a meme. Analyzing the meme, both the word content and picture, together presents a challenge to developers.
A meme can be harmless or hateful simply by switching the imagery behind the text. Facebook wants AI that can differentiate the alternative meanings of a statement on two varying backgrounds. The company labels hateful speech as content that is directly or indirectly attacking characteristics such as ethnicity, race, nationality, immigration status, religion, caste, sex, gender identity, sexual orientation and disability or disease. Mocking hate crime is also considered to be hate speech in the eyes of Facebook. Therefore, the company looks to award a developer that can produce AI that has a hate speech identification success rate of 84.7%, similar to that of a human. Participants have until October to deliver AI that has the capability of distinguishing hateful memes from a dataset derived from existing memes.