Facebook comes up with "Hateful meme challenge" to improve its AI

Ojaswin Kathuria
Thursday, 28 May 2020

The current social media environment is highly volatile, thus such algorithms are needed to combat hate-speech.

Facebook's artificial intelligence team on May 12 announced the launch of a challenge which aims at deleting out or recognising hateful meme content on the social networking platform. The challenge is called "Hateful meme challenge" and is hosted by DrivenData with US $100K prize pool.

The artificial intelligence team has also created a Hateful meme data set which has 10,000+ which has multimodal examples to help researchers and the participants of the competition to come up with an algorithm that can detect hateful memes.

Memes are a representation of a context using images and words. Thus, meme content combines different modalities, such as text and images, making it difficult for machines to understand. To make artificial intelligence more effective a crucial challenge is to make such algorithms, so that AI understands the content just like the human brain does, i.e. holistically.

To give perspective, when we look at a meme, we don't see the text and the image as independent of each other, we understand the meaning and the context combined. This is extremely difficult for machines and AI models because they cannot analyse the text and image as one whole context. Facebook AI has created a data set to catalyse research in this area to help build systems that better understand multimodal hate speech.

Facebook has released this data from Hateful Memes to help the broader research community and also launched an associated competition, hosted by DrivenData with a prize pool of US $100,000.

Models like these are extremely important in the current highly volatile social media environment, which is full of hate speech; the Facebook team defines hate speech as:

"A direct or indirect attack on people based on characteristics, including ethnicity, race, nationality, immigration status, religion, caste, sex, gender identity, sexual orientation, and disability or disease. We define attack as violent or dehumanising (comparing people to non-human things, e.g. animals) speech, statements of inferiority, and calls for exclusion or segregation. Mocking hate crime is also considered hate speech."

This definition reflects the collective guidelines on hate speech used by Facebook and is meant to include an actionable classification label: if anything is hate speech in accordance with this definition, it will be removed; if not, even though it is distasteful or offensive, it is permitted to stay.

You can find more information related to the challenge at DrivenData

Also Read: Sudhir Krishnaswamy the only Indian on Facebook's Oversight Board

Related News

​ ​