AI Deployed to Combat Anti-Polish Hate Speech

Dominique Adams

08 August 2019, 12.45am

Artificial Intelligence is being used to tackle the rising problem of anti-Polish hate speech on social media.

HateLab, a global hub for insight into hate speech based at Cardiff University, has partnered with Polish company Samurai Labs to deploy its artificial intelligence (AI) tech to monitor aggressive social media content.

As well as monitoring this type of content, the technology will also seek to identify connections between online content and offline events.

In a effort to pre-empt outbreaks of hate crime on the streets, HateLab developed its Online Hate Speech Dashboard in partnership with academics and policy partners. It is powered by AI algorithms developed by Samurai Labs.

Between 2017 and 2018, more than 94,000 hate crimes were recorded in England and Wales – an increase of 17%. Research by HateLab found that race hate crimes represent 76% of all offences, followed by sexual orientation (12%), religion (9%), disability (8%) and transgender (2%) hate crimes.

More than 900,000 Polish people live in the UK, making them Britain’s largest minority. Research carried out by Samurai Labs found that 5% of material published online about Polish people had negative or offensive connotations.

According to researchers at Cardiff University, AI could be used to “accurately differentiate between web aggression and harmless comments”. Samurai Labs chief executive, Michel Wrozynski, said AI is capable of identifying different categories of abuse, from people being told to go home to death threats.

Matthew Williams, director of HateLab, said this technology is vital due to the surge in hate speech since the 2016 EU referendum. “As the United Kingdom prepares to leave the European Union, using the most advanced methods of AI is going to be vital in helping the authorities to quickly recognise warning signs and provide reassurance and security to the Polish community living here,” he said.

Gniewosz Leliwa, director for AI Research at Samurai Labs, said: “We have to be very precise and work with a deep understanding of the language. Keywords are certainly insufficient. It is possible for someone to make a punishable threat without using any of the characteristic words.

“Similarly, the use of vulgarity is not always necessarily associated with an aggressive message. Our system is able to understand the context and correctly separate a real threat from an ordinary conversation.”