Africa Check, Google.org partner to fight false information using AI

Fact-checking organisations Africa Check, Full Fact and Chequeado received a $2million (Ksh. 218,700,000) grant alongside seven Google.org Fellows who spent six months building a tool which uses artificial intelligence (AI) to fight misinformation.

All three leading fact-checking organizations now use AI to provide trend monitoring and clustering tools to aid fact checkers’ analyses of news and other information.

Kate Wilkinson, Deputy Chief Editor at Africa Check said: “False and misleading information threatens us all. 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.”

“The tool helped us identify false statements made in South Africa by the head of our power utility and the finance minister. Without the AI technology, the incorrect claims would have gone unchallenged and misled the public,” she added.

During the six-month fellowship, the Google.org Fellows built machine learning tools to identify checkable “claims” in news media (who, what & when) as concerns about mis- and disinformation grow.

This follows the huge amount of misinformation unleashed during the COVID-19 pandemic; the most recent trend being vaccine misinformation.

According to Africa Check, bad information ruins lives: it damages people’s health, leads to poor decision making and undermines democracy.

Given the massive amount of online media produced daily, Africa Check said its team of expert fact-checkers find it impossible to scan it all to find the most important claims to fact-check.

For those reasons, the Africa Check team along with its partners Full Fact and Chequeado explored whether AI could help identify which are the most important claims in the news media and radio to prove or disprove.

Alexios Mantzarlis, News and Information Credibility Lead at Google said: “Too much time and money has been spent chasing a silver bullet end-to-end automated fact-checking solution. The reality is that fact-checking is highly contextual and will always require humans in the loop. The coalition led by Full Fact has shown that automation can be extremely valuable if placed at the service of fact-checkers, accelerating the work of discovery and concentrating valuable time on the actual verification. It is this type of innovation, rooted in newsroom needs and processes, that the field needs to make progress.”

The AI tools are said to have increased claim detection across the three organisations 1000 times; from several hundred claims to 100,000 per day.

The Google.org Fellows also built a natural language processing model to link claims related to the same person who might be referred to in different ways: (e.g. Cyril Ramaphosa, the President, Ramaphosa, or even “Cyril”).

Additional outputs included a speech to text model to match claims made on radio with claims made in text media; a search engine to scan thousands of articles to find similar claims that someone has made even if they use different terms (e.g. R1B, R1 billion, R1,000,000,000) or verb tenses as well as a web application to highlight claims that need to be checked, and provide links and tools for a journalist to start researching the claim’s validity.

Gabriel Rovina, Google.org Fellow & Machine Learning Engineer said: “It hit close to home when I opened my family chat group and saw my immediate family members sharing fake news and inaccurate information about candidates and voting. I knew right then that I had to take action and dedicate time to become part of the solution”.

Africa Check said the BERT based model works across 4 languages (English, Spanish, Portuguese and French) and the intuitive interface developed with Google has led directly to fact-checks being published in South Africa, Nigeria, Kenya, Argentina and the UK.

In 2020, over 8 million people are reported to have visited the Africa Check website in search of accurate, evidence-based information.

Fact-checks are said to have appeared in 9.3 million search results and of those 1.62 million times as a rich result where they were specially shown as a fact-check (see below).

Africa Check said the AI tool has allowed them to fact-check incorrect statements that might have gone unchallenged such as when the CEO of South Africa’s public electricity utility Eskom wrongly claimed that blackouts during the country’s Covid-19 lockdown were caused by the ‘coldest winter in 10 years’. Historical weather data proved him wrong.

South Africa’s minister of finance, Tito Mboweni, also claimed ‘almost 100%’ of restaurant workers in the country were ‘non-South African’. Official government data disapproved his statement.

Jen Carter, Founder of the Google.org Fellowship Program said: “We created the Fellowship Program to bring the best of Google to incredible nonprofits like Full Fact, Chequeado and Africa Check. We were thrilled to be able to support these organizations’ expert fact checkers by building AI-powered tools to help them be more efficient & effective in their jobs.”

On his part, Cong Yu, Team Lead, Google Fact-Checking said: “The technology built by Full Fact, Africa Check, Chequeado and the team of Google.org Fellows is an example of incredible global collaboration to tackle one of the world’s hardest problems. The fact that the AI model has boosted the number of detected claims by 1000x across 4 languages, and 3 continents is very impressive.”

And Rowan Barnett, Head of Google.org EMEA & APAC said: “Tackling misinformation is a critical challenge, and by leveraging AI and technology, working with innovative organisations on the ground such as Africa Check and Chequeado, and with the support of Google.org Fellows and resources, we’re delighted to have helped Full Fact make significant progress in tackling it.”