Pioneering Social and Behavior Change with Generative AI

This year’s first session of the Generative AI (GenAI) and Social and Behavior Change (SBC) Working Group was organized by co-leads Sarah Osman, Nicola Harford, and me (Stephanie Coker). At the session we explored the applications and implications of GenAI within social and behavior change program design and implementation, particularly for international development programs.

Our three speakers, Isabelle Amazon-Brown, a digital development consultant specializing in human-centered design and digital social impact products working with The MERL Tech Initiative and iMedia Associates; Lukas Borkowski, Senior Director of Strategic Partnerships at Viamo; and Oluwaseun (Seun) Nifemi, Lead Data Scientist and AI engineer at Data Science Nigeria (DSN), focused on how Gen AI can be effectively integrated into SBC programming.

Is GenAI is Revolutionizing SBC Programs?

Isabelle kicked off the session by sharing back insights from a workshop she co-facilitated at the ICT4D Conference in Accra in March 2024. With 50 practitioners in attendance, this workshop explored whether GenAI was really going revolutionize SBC programming. It highlighted potential uses of GenAI in areas such as fundraising, program design, monitoring, and evaluation. Despite the excitement around GenAI, most of the workshop attendees shared that they are still in the early stages of using tools like ChatGPT and need significant capacity building and guidance to effectively integrate these technologies into their work.  Transparency in sharing performance data and early-stage results from case studies will be crucial for creating sensible benchmarks and not reinventing the wheel.

Applications of Gen AI for Social and Behavior Change Programs

Lukas spoke about Viamo’s work using Gen AI to deliver personalized advice through voice-first approaches. Voice is of particular importance for populations not yet online. He demonstrated how Viamo’s platform leverages GenAI to provide contextualized, localized information via basic non-internet phones. Lukas also discussed the significant potential of Gen AI to leapfrog traditional technologies, providing highly personalized services without the need for smartphones or internet connectivity. He highlighted the importance of transparency, control over content, and real-time data reporting to measure effectiveness and adapt programming.

He discussed Viamo’s use of Retrieval Augmented Generation (RAG), a GenAI technique that enhances the quality of responses by retrieving relevant information from a content store before generating an answer. Viamo uses RAG to generate responses from existing, approved documents provided by domain experts. These responses are then conversationalized by a large language model (LLM) like GPT-4. RAG improves relevance and accuracy because it draws from data that is specific to the theme or domain (e.g., health) rather than directly generating responses through an LLM which is trained on all documents on the internet.

Viamo, uses RAG to quickly update their voice-first GenAI with additional documents loaded into the content store, a process that is easy and cost-effective (and actionable for non-tech organizations) as compared to re-training an entire LLM. Viamo’s presentation also included a live demo, showcasing how users can interact with their GenAI-powered voice service. 

Seun presented Data Science Nigeria’s work on leveraging Gen AI for sexual and reproductive health behavior change. DSN developed a model that assesses user readiness for behavior change and provides personalized, context-specific information on uptake and encourages continuation of contraceptive use. Seun underscored the need for emotional intelligence in AI and the potential for AI-augmented decision-making to support strategic planning processes. She also highlighted the importance of user feedback to measure the impact and effectiveness of Gen AI interventions.

Question & Answers with our speakers

We had a lively Question & Answer segment where, participants highlighted the importance of transparent sharing of case studies and data to benchmark GenAI’s cost and effectiveness in SBC. 

For example one person asked how much it costs to deploy this type of service in a specific country and the variables that should be considered that feed into the cost. Lukas shared that “The voice-first GenAI Viamo Ask an Expert (AAE) pilot costs USD 100k for 12 months in one of the 22 countries in Africa and Asia where we have the existing Viamo Platform (toll-free national-scale SBC hotline). This covers one pilot, one country, one language, one domain. We take 6 months to develop the pilot and it is then live for 6 months on the Viamo Platform.”

Another session participant asked DSN how they personalize responses based on user characteristics. Seun explained that responses are based on users questions. “Because the language model is fine tuned with local nuances, it is able to recognize the language style of communication and give personalized responses.”

Other people raised concerns about user readiness and ethical implications, stressing the need for informed consent and addressing digital literacy inequalities. Questions were also asked about the potential of GenAI to close the gender digital divide and provide real-time, contextual information during health crises like cholera outbreaks. Emphasis was placed on understanding local context and language in AI solution design.

(You can find detailed responses to participant questions here!)

Developing a shared research agenda on GenAI and SBC

Nicola led facilitated discussion on creating a shared research agenda for SBC programs, which further illuminated several key points:

• The need for rigorous impact evaluations to test the hypothesis that Gen AI can supercharge SBC by delivering targeted, customized information at scale.
• The importance of sharing both successes and failures to avoid repeating mistakes and to build a robust evidence base.
• The necessity of involving human experts when dealing with sensitive topics and ensuring that AI tools have appropriate guardrails.
• The need for donor and stakeholder education to alleviate fears and misconceptions about AI, promoting a better understanding of its potential and limitations.

Overall, the session highlighted the exciting possibilities of Gen AI for SBC while also emphasizing the need for cautious, well-informed implementation and continuous learning. By sharing knowledge, experiences, and data, the NLP SBC working group will continue to collectively navigate the challenges and opportunities presented by Gen AI in the SBC field.

Wrapping up and next steps

• Details for our next NLP CoP SBC meeting are forthcoming.
• To get in touch with event speakers Lukas and Seun, fill out this contact form.
• For more information on Gen AI tools for SBC,  check out the Development & Humanitarian GenAI tools directory compiled by Issie.
• Join us in the NLP CoP,  a community of over 700 development and humanitarian practitioners convened by The MERL Tech Initiative if you’d like to stay up to date on GenAI and NLP or to get information about future meetings on SBC and other topics!

We need funding to enable the SBC Working Group and its members to respond to the various questions raised (both practical and theoretical). Learn more about how you can support the NLP-CoP and its working groups. If you are interested in sponsoring or supporting this work, please get in touch – we’d love to discuss this!