Tama Media, a Pan-African media outlet that employs African journalists to tell local stories, has developed an innovative AI-powered chatbot designed to combat misinformation while breaking down language barriers. The organisation's approach demonstrates how artificial intelligence can be adapted to serve local contexts, particularly in regions where traditional fact-checking methods struggle to reach vulnerable populations.

The problem: Reaching beyond digital divides

Africa faces a unique challenge in the fight against misinformation. As Tama Media’s project lead Moïse Mounkoro explains: “In Africa, there are people who do not or cannot read, and yet, they still use social media platforms where they easily can share fake news without being aware of it”. Due to this, traditional fact-checking approaches – articles, videos, and podcasts – aren’t effectively reaching these communities.

Tama Media recognised three critical challenges: making fact-checking easier and more engaging; ensuring accessibility for people who primarily communicate in local African languages rather than French or English; and reaching illiterate social media users.

"The idea is to see how AI can contribute to fighting misinformation – making fact-checking easy, fun, and accessible."

Building the solution: AI meets local expertise

Tama Media’s response was to build an AI-powered chatbot that users can interact with through voice or text. The chatbot allows users to verify information by simply asking questions like, "I have seen this on social media, is it true?" The system responds with verification and provides multiple sources to support its answers. Crucially, the tool integrates with popular platforms like WhatsApp and Facebook, meeting users where they already are.

What sets this project apart is its hybrid approach. The team built their solution using both ChatGPT and a custom-trained AI model for local African languages, starting with Bambara – a language spoken in Mali and, in related forms, across West Africa, particularly in Côte d’Ivoire, Burkina Faso and Republic of Guinea.

The system maintains quality through careful source curation. "We selected media news outlets that are known and that we are sure don't spread fake news,” explains Mounkoro. "We have selected outlets including BBC Africa, RFI, France 24, and local African news outlets."

When users aren't satisfied with responses, human oversight kicks in. "When the user responds negatively, the system sends our newsroom a message that there has been a question and the person is not satisfied. When we answer the question, the user gets an alert that the answer is ready."

Building local capacity

The project's development involved two distinct teams: journalists responsible for data collection and user responses, and a technical team based in Abidjan. The technical team includes specialists in AI, design, and integration – a deliberately local approach that initially faced scepticism.

"I talked to several experts in Paris about this tool and they said that it’s too complicated and unfeasible. But then we got in touch with a technical team in Côte d’Ivoire, they acknowledged its complexity but they rose to the challenge."

This local focus proved crucial for creating a tool designed by Africans for African users, despite the technical challenges involved.

Overcoming technical hurdles

The team encountered several significant challenges. Initially, the system took five to 10 minutes to process queries – impractical for social media users expecting instant responses. Through persistent optimisation, they reduced response times to 20 - 30 seconds.

Accuracy presented another hurdle. The team needed to ensure 80 - 90% reliability in fact-checking results while working with limited resources as a small, local team.

Perhaps most significantly, integrating African languages into AI systems designed primarily for Western languages required innovative solutions. The team collected extensive data in Bambara – podcasts, dictionaries, and other materials – to train their custom model.

Their solution involves real-time translation: when users submit queries in local languages, the system translates them to English or French for processing, then translates responses back to the original language for delivery.

The opportunities: Partnerships and sustainability

The team plans to offer their AI model to other African newsrooms, enabling similar fact-checking capabilities across the continent. They're also exploring sustainable revenue models through advertising that could support ongoing fact-checker compensation.

They also envision their project as a foundation for broader collaboration. “We hope to expand future partnerships to include non-media organisations in fields affected by misinformation, such as the World Health Organization.”

Lessons for newsrooms

Tama Media's experience offers valuable insights for similar initiatives worldwide. 

• Build on existing strengths: Launch AI projects by starting with your team's existing expertise in areas like fact-checking. This provides a strong, practical foundation for success.

• Set realistic AI expectations: Remember that AI is a tool, not a "magic" solution. It requires study and patience. It's crucial to understand its limitations and what problems it can and cannot solve for your organisation.

• Integrate human and machine efficiency: Effectively fighting misinformation in diverse languages requires combining AI's efficiency with human expertise and local knowledge. This creates a scalable model that respects cultural and linguistic nuances.

• Prioritise symbiotic collaboration: The future of fact-checking is about creating a symbiotic relationship where AI amplifies human capabilities. Don't aim to replace human experts, but rather to use technology to enhance their local knowledge and better serve the community.