Language technologies like generative artificial intelligence (AI) hold significant potential for public health. From outbreak detection systems that scan global news in real time, to chatbots providing mental health support and conversational diagnostic tools improving access to primary care, these innovations are helping address health challenges.

At the heart of these developments is natural language processing, an interdisciplinary field within AI research. It enables computers to interpret, understand and generate human language, bridging the gap between humans and machines. Natural language processing can process and analyse enormous volumes of health data, far more than humans could ever handle manually. This is especially valuable in regions with a stretched healthcare workforce or limited public health surveillance infrastructure, because it enables faster, data-driven responses to public health needs.

Recently, our interdisciplinary team, combining expertise from computer science, human geography and health sciences, conducted a review of studies on how language AI is being used for public health in African countries. Almost a decade’s worth of academic research was analysed, to understand how this powerful technology is being applied to pressing human needs.

Out of 54 research publications, we found that evidence of real-world effects of the technology was still rare. Only 4% of these studies (two out of 54) showed measurable improvements in public health, such as boosting people’s mood or increasing vaccine intentions.

Most projects stop at technology development and publication. Very few advance to real-world use or impact. Opportunities to improve health and well-being across the continent could be missed as a result.

Current limitations

In recent years, AI language technologies for public health have increased rapidly. This wave of technology development really took off as the COVID-19 pandemic renewed attention to public health. Health chatbots and sentiment analysis tools were developed in Africa and beyond.

Graph showing lines — Research on language AI for public health in Africa. Supplied

Health chatbots “talk” to people and provide reliable health information in a friendly, conversational way. Sentiment analysis tools scan social media posts to understand what people are feeling and talking about. Together they can identify misinformation or changes in public opinion and then provide accurate information.

Of course, new technologies come with imperfections. We found that most technologies for public health in Africa exist in just a few languages whose dominance can be traced to colonial times, namely English and French.

The consequences are clear: key health messages fail to reach many communities, leaving millions unable to access or act on essential information.

We also found that few projects have gone beyond the laboratory development stage. Our study found only one system in operation that had a measurable public health effect.

A successful model

This standout example comes from a team at the Center for Global Development and the University of Chicago, in partnership with the Busara Center for Behavioral Economics. Their chatbot, deployed on Facebook Messenger, was designed for people in Kenya and Nigeria who were hesitant about COVID-19 vaccines. It was only available in English.

More than 22,000 social media users used this app, sharing vaccine-related questions and concerns. The chatbot provided tailored, evidence-based responses to topics ranging from vaccine effectiveness and safety to misinformation. Its effect was notable. The intervention boosted users’ intention and willingness to get vaccinated by 4%-5%. The strongest effects were seen among those most hesitant to begin with.

Behind this success was the researchers’ commitment to understanding the local context. Before launching the chatbot, in-depth discussions were held with focus groups and social media users in Kenya and Nigeria. The aim was to learn about the specific worries and cultural factors shaping attitudes toward vaccination.

The chatbot was designed to address these concerns. This user-centred, locally adapted approach enabled the chatbot’s messages to address real barriers. As this example demonstrates, language technologies for public health are most effective when responding to the concerns and needs of the intended users.

From lab to life

These technologies take time and money to be put into practice. The COVID-19 pandemic jump-started development but public health language AI technologies are very new. It could be that a future survey would find a very different situation.

At the same time, advances in large language models such as GPT-4 are rapidly lowering the technical barriers to developing language technologies. These models can often be adapted to new applications with far less data and effort than previous methods. Recent advances could enable small teams of researchers or even individual developers to build tools tailored to the specific needs of their own communities. The path from lab to real-world effects may become much shorter and easier.

Investors, accelerators and state support could help make this transition from lab to life happen.

Technology developers can also contribute by rooting their work in community-driven, multi-disciplinary and cross-sector collaboration. Social science and public health research knowledge and skills can inform the design and development of new technologies.

To maximise the potential of language technologies for public health, the following needs to happen:

involving communities and health workers in natural language processing design
expanding provision in indigenous African languages
integrating language technologies into existing health systems.

Future research and development must move beyond technical prototypes and laboratory tests to rigorous real-world evaluations that measure health outcomes.

The other co-authors behind this research are: Abigail Oppong, Ebele Mogo, Charlotte Collins, and Giulia Occhini.