When misinformation spreads online, speed is everything. A single false claim, launched into the digital world during the chaos of an election, the panic of a public emergency, or the uncertainty of a health crisis, can spiral out of control before anyone has a chance to correct it. It shapes decisions in real-time, fuels confusion, and can put entire communities at risk. The danger is immediate and visceral—a rumor can stop someone from getting vaccinated, incite panic buying that empties store shelves, or provoke violence based on a fabricated story. Yet, in this high-stakes race to separate fact from fiction, the tools we rely on are often not built for everyone. The vast majority of artificial intelligence systems developed to detect and flag misinformation are designed with English as their primary language. This leaves the hundreds of millions of people who communicate in other languages, particularly those in multilingual nations like South Africa, navigating a digital landscape with significantly fewer safety nets. For a speaker of isiZulu or Sepedi, the protection against a viral lie is often thinner, the response times slower, and the resources for verification scarcer, simply because the technology was never designed with their language in mind. This is not just a minor oversight—it is a digital vulnerability that affects real lives.
This is the crucial gap that the North-West University (NWU) is stepping in to address. Recognising that technological progress must serve everyone, not just a privileged few, the University has initiated a vital research project focused on building multilingual AI that can effectively identify misinformation across English, isiZulu, and Sepedi. It is a move that acknowledges a simple truth: for AI to be truly useful in South Africa, it must speak the languages of South Africa. By building this framework, NWU is working to level the digital playing field, ensuring that one’s mother tongue is not a barrier to accessing truthful information. The initiative is more than a technical exercise; it is a commitment to digital equity. It is about creating a technological future where the rich linguistic diversity of the country is not seen as a complicating obstacle to be worked around, but as a core design principle to be embraced. In this vision, an AI tool is not considered complete or effective unless it can seamlessly understand and analyse the nuances of a Sepedi message just as well as an English one. This approach signals a shift from a one-size-fits-all technology model to one that is deeply embedded in the local context, respecting and protecting the voices that make up the South African story.
To test their framework, the researchers at NWU turned to a recent, painful shared experience: the COVID-19 pandemic. This global crisis was accompanied by an equally global “infodemic,” a torrent of misinformation that proved to be nearly as dangerous as the virus itself. From false cures and dangerous health advice to conspiracy theories about the origin of the disease, misinformation spread across every corner of the internet, often finding fertile ground in communities where authoritative, local-language information was scarce. This made the pandemic a powerful pilot case study for the multilingual AI framework. The research team put their system to the test, seeing if it could accurately sift through the digital noise and identify misleading information across English, isiZulu, and Sepedi simultaneously. The results are promising, showing that a single, cohesive framework can be trained to spot falsehoods in multiple, structurally different languages. The work also delved into a critical technical challenge: the importance of enhancing machine translation for these so-called “low-resource” African languages. In the world of AI, some languages have vast amounts of digital text to learn from, while others have very little. This scarcity can make translation and understanding less accurate, which significantly hampers the performance of AI tools. The NWU research highlights that improving the translation quality for isiZulu and Sepedi is not a separate project but a fundamental pillar in building a stronger, more reliable overarching detection system. The better the translation, the better the AI can understand the true meaning behind the words, and the more effectively it can flag misleading content.
The core purpose of this work is deeply humanistic: to ensure that speakers of indigenous languages are not left behind in the age of artificial intelligence. As AI becomes more integrated into our daily lives, particularly in the fight against online misinformation, it has the potential to either bridge the digital divide or deepen it. Without tools like the one being developed by NWU, speakers of isiZulu and Sepedi are left more exposed to the dangers of false information. They are forced to rely on translation, which can be slow and inaccurate, or to simply navigate a digital world where they are more vulnerable. This project is about empowering these communities, giving them the same digital resilience as English speakers. The potential applications are vast and meaningful. Government departments could use this framework to monitor public sentiment and dispel dangerous rumors more quickly, ensuring that official communication reaches citizens in a language they understand. Fact-checking organizations could expand their crucial work beyond the limits of English, verifying claims that circulate in local languages and providing clarity to a broader audience. Social media platforms could integrate this technology to monitor and moderate content on their South African user base more effectively, creating a safer online space for everyone. Ultimately, for individuals, this means earlier detection of fake news, greater access to credible information, and the ability to make informed decisions about their health, their vote, and their community, without being swayed by malicious or false narratives.
The vision for this research is inherently long-term and expansive. While the initial success with COVID-19 provides a strong foundation, the researchers at NWU are already looking ahead to the next set of challenges and opportunities. The next phase will involve expanding the framework to encompass more of South Africa’s twelve official languages, a monumental but necessary undertaking to ensure comprehensive national coverage. Beyond just general misinformation, the focus will sharpen on high-impact social media content during crucial periods, such as elections, where misinformation can undermine democratic processes, and public emergencies, where the accuracy of information can be a matter of life and death. The research will also need to evolve to counter a new and more insidious threat: AI-generated deepfakes. These hyper-realistic manipulated videos and audio recordings are becoming easier to create and harder to detect, posing a massive challenge to the integrity of online information. Future work will therefore explore how the multilingual framework can be adapted to identify deepfakes, track hate speech, and monitor other forms of harmful online content that disproportionately affect marginalized groups. This forward-thinking approach ensures the research remains not just relevant, but essential, in the ever-changing landscape of digital communication.
The potential long-term impact of this foundational work is to contribute to the development of truly trustworthy and inclusive AI systems. This goes far beyond just a fact-checking tool. The research paves the way for creating sophisticated large language models and knowledge graphs that are fluent in South African languages, not just as a translation exercise, but as a means to understand and generate culturally relevant and contextually accurate information. These systems could become invaluable assets in a wide range of sectors, transforming how services are delivered and accessed. In healthcare, such an AI could help community health workers in a rural clinic navigate a complex medical database in Sepedi. In education, it could serve as a tutor for students, explaining concepts in their home language. In agriculture, it could provide farmers with real-time market data or weather forecasts in isiZulu. In governance, it could help citizens understand their rights, access public services, and engage with government in their own language. The work at NWU is a perfect example of how research that begins with a clear purpose can blossom into knowledge that serves society in profound and lasting ways. By tackling this challenge head-on, NWU is not just building an algorithm; it is building a bridge to a future where technology respects and reflects the full, vibrant diversity of the nation it is meant to serve. It ensures that the power of AI to protect against falsehood is a right shared by all South Africans, not a privilege for the few.