Fighting fake news with AI â how west Africaâs Dubawa is transforming fact-checking

This story has significance for readers across Africa and beyond.

In their modest office in Abuja, Nigeria, Monsur Hussain and his team of software developers and AI researchers are debating ways to improve the next generation of fact-checking technology. Laptop screens flash with lines of code and machine-learning models while the group compares test results and revised algorithms, demonstrating how AI could better identify fact from fiction.

Hussain is head of innovation at the Centre for Journalism Innovation and Development (CJID), a west African media innovation and development think-tank that uses investigative journalism, research and technology to improve journalism, promote accountability and combat misinformation across the region.

He led the development of Dubawa AI, a fact-checking tool launched in 2024 to help journalists and the public evaluate claims before they spread online.

The team began by focusing on one challenge: teaching AI to detect factual claims that needed to be verified.

âThereâs a difference between questions and claims. What we want to focus on is claims,â Hussain says. âWe started with training a model on understanding what claims are, especially claims that have an impact on society, like health-related claims, political claims and others.â

The Nigerian Communications Commission reports that Nigeria has more than 148 million active internet subscriptions, making it Africaâs largest online market. At the same time, advancements in generative AI have made modified photos, videos and audio more convincing, putting more pressure on journalists to check content before it is released.

Regional outlook

For Abuja-based software engineer Japhet Johnson, locally developed AI models are increasingly important because they are trained using regional information that many global models overlook.

âMost of these AI models are trained maybe for Europe or the United States,â he says. âWhen we want to actually prove or get information, we find that most of those AI [models] might hallucinate. They might not be able to give you good information about Africa. That is where Nigeria needs to step in. We need AI that understands how our news is being presented.â

Johnson believes strengthening local data infrastructure will be critical if African AI systems are to remain reliable while complying with national data protection laws. âWhat we need to do now is ensure we have companies that follow the Nigeria Data Protection Regulation rules so that a lot of our data is stored internally.â

Chatbot for fact-checking

The team created a chatbot for the Dubawa website and WhatsApp that allows users to check claims in seconds. It also has an audio tool that pulls claims from recordings for fact-checking.

âIf you go to dubawa.org, you will find the Letâs Factcheck! icon on the bottom right corner,â Hussain says. âYou put in the claim, click submit, and itâs able to give you the verdict on that particular claim.â

Hussain says the chatbot has verified more than 6,000 claims and has been used in more than 32 African countries, with growing adoption in Ghana, Gambia, Liberia, Sierra Leone and Senegal.

To train the algorithm, the developers linked years of verified Dubawa fact-checks to a machine-learning model that can detect trends in previously reviewed allegations. âWe built a website with Python and connected Dubawaâs database where all the fact-checks were stored,â Hussain says.

Instead of launching the platform right away, the team created a minimum viable product and tested it with journalists and fact-checkers, polishing it through regular feedback. âWe got feedback and then did a lot of iterations until we got something that was more satisfactory.â

The research emphasises the magnitude of the challenge. A study titled Fake News and Public Trust in Online Media: A Study of Abuja, Nigeria discovered that 75% of internet users polled frequently encountered fake news on their principal social media platforms.

Saving time

For journalists, verifying fast-moving claims frequently entails exploring different databases, tracking down original sources and cross-checking evidence before publishing. Hussain believes that AI may shorten the time spent on those jobs while preserving editorial judgement.

âItâs about trying to reduce the misinformation that flies everywhere and trying to monitor what people are saying, what claims people are making and how we can put a control on it so that people understand thereâs a form of accountability,â he says.

For Ebube Glory Nwankwo, a media researcher at the CJID, the chatbot has changed how she monitors information gathered from social media and other online platforms.

âMy work as a researcher involves a lot of media monitoring and a lot of information on social media,â she says. âDoing it manually before Dubawa AI actually takes a lot of time, days sometimes.â

The ability to verify information on a cellphone has also changed how she works while away from her desk. âI donât have to bring my laptop to check for information or sources. If I take my phone, Iâm able to do it, and that has really made my life easier.â

Because AI models can create false data, the team implemented protections before making the chatbot public. âWe implemented certain changes and put in guardrails to ensure the model does not go out of context in a lot of situations,â Hussain says.

For Hussain, however, the platformâs success depends less on the number of people using it than on whether it changes how people consume and share information.

âI think what we want to build is a culture where people learn to check before they forward information, before they send something out.â â Bird Story Agency/DM

This story first appeared in our weekly DM168 newspaper, available countrywide for R35.

Reporting originally appeared via Daily Maverick. Read the full source for additional context.