On 2 April 2021, the global fact-checking community again observes International Fact-Checking Day. It’s an annual celebration for our community and a rallying cry for better information in public health, journalism and everyday life.
April 2021 also marks a year of Covid-19 lockdown. Around this time last year, countries began introducing restrictions on the movement of people in order to slow the spread of the disease. It was also around this time last year that Africa Check staff packed up their work bags and headed home, unsure of when we’d be able to return.
Spreading almost as fast as the pandemic has been what the World Health Organization calls an infodemic about the disease.
Covid-19 misinformation was one of the year’s toughest challenges. It exposed a danger fact-checkers have warned about for years: false and misleading claims can ruin lives. .
And it highlighted a question long debated by fact-checkers. Hundreds of harmful claims are made every day. How do we know we’re checking the most important ones?
We get by with a little help from our (AI) friends
In 2019, Africa Check and three of its international partners received a US$2 million grant from Google.org to use machine learning to dramatically improve and scale global fact-checking efforts.
Over six months, and with the help of six Google.org fellows, we’ve built artificial intelligence tools to detect claims made by people in the public eye, group them by topic, and match them with similar claims in both traditional and social media.
Rather than spending hours sifting through media content for possible claims, fact-checkers can now directly access a list of statements AI has already identified as being fact-checkable. This helps us to respond more efficiently to misinformation. It also alerts us to important claims we might otherwise have missed.
Electricity supply and migrant workers
Here are a few examples.
In October 2020 André de Ruyter, CEO of South Africa’s government-owned electricity producer Eskom, claimed that blackouts across the country had been caused by “the coldest winter in 10 years’.
Although temperatures that winter were found to be generally below normal, data did not support the claim. And experts said the power system should be able to handle winter demand peaks.
Our AI tools also picked up a startling claim by South Africa’s minister of finance, Tito Mboweni.
When he returned from exile in 1990, he said, eight in 10 restaurant workers were South African. “The other two were probably Malawian or Zimbabwean. Today almost 100% are non-South African.” Again, no data supported the claim. In a country where tensions between citizens and foreign nationals often run high, unsupported claims like this do little to help.
We’ve also published fact-checks on youth unemployment, mental health and transport, all on claims identified by AI.
Watch this space
The AI is not designed to replace fact-checkers. It allows us to spend more time doing the work that matters most – checking false and misleading claims.
Using a machine learning BERT-based model, the technology now works in four languages – English, Spanish, Portuguese and French – and across five countries.
“Google.org’s support has allowed us to build tools that make fact-checkers’ work faster and more effective – whether you’re based in London, Lagos, Buenos Aires, Nairobi or Johannesburg,” said Kate Wilkinson, deputy chief editor at Africa Check.
By amplifying our work, AI is helping millions of people access accurate information on a broad range of topics. And according to Full Fact CEO Will Moy, this is just the beginning. “I am pushing [to make] these amazing tools available to all of our fact checkers and in time around the world,” he said.
Add new comment