A cross-disciplinary team from The Alan Turing Institute and The University of Edinburgh has placed in the top 1% of more than three thousand teams who registered for the Hateful Memes Challenge. The challenge, jointly organised by Facebook AI Research and DrivenData, was designed to create an algorithm that can identify whether a meme - consisting of an image and overlaid short text - is either hateful or benign.

Social media has been transformative in allowing people to quickly share ideas and content. Unfortunately, this has also led to some groups of people being targeted online with hateful content. Online hate risks harming vulnerable groups and communities, toxifying public discourse and exacerbating social tensions. Pressure is growing on major platforms to remove such content. However, the sheer volume of online posts makes manual moderation of social media conversations near impossible, necessitating the use of automated tools.

These automated hate detection systems are much less effective when toxicity arises from the combination of an image and text. Such “memes”, which cause offence through the interaction between images and text, remain particularly difficult to detect using even the latest AI technologies. Most algorithms are often trained on one modality i.e. just text or just images. To tackle this, the team developed a new algorithm capable of understanding image-text pairs. It outperforms previous state-of-the-art algorithms for hate detection and can be applied to other tasks involving vision and language inputs. The algorithm created by the team is being refined further and will be presented at international conferences in the near future.

Grzegorz Jacenków, a PhD student at The University of Edinburgh and the project lead, reports "I am thrilled to see our method contribute to the challenge of fighting harmful content online. Humans have an astonishing capability to interpret multimodal interactions effortlessly while AI algorithms struggle. We need to push AI algorithms towards a holistic understanding of the world - this is the topic of my doctorate research, with a focus on the medical domain."

"I am thrilled to see our method contribute to the challenge of fighting harmful content online. Humans have an astonishing capability to interpret multimodal interactions effortlessly while AI algorithms struggle. We need to push AI algorithms towards a holistic understanding of the world - this is the topic of my doctorate research, with a focus on the medical domain."

The team was generously supported with Azure server credits by the Turing and given access to JADE, the largest GPU facility in the UK funded by the EPSRC for world-leading research in machine learning.

The team consists of Grzegorz Jacenków, Elena Kochkina, Harish Tayyar Madabushi, Bertie Vidgen, Maria Liakata, Helen Margetts, Alison O’Neil and Sotirios Tsaftaris.