AI at APHRC: How Are We Using It?
- Where do you think AI could have the biggest impact at APHRC?
- What AI tool have you tried, and what task did it help you accomplish?
One AI use case I would like to explore is using AI to analyse adolescent health and wellbeing data to identify patterns and gaps in areas such as menstrual health, teenage pregnancy, school nutrition, and access to mental health services. This could help APHRC generate timely, evidence-informed recommendations while ensuring that data privacy, ethical considerations, and African contextual knowledge remain central to the proce
AI is very crucial currently at APHRC since it can be applied to control large population and educational data from the surveys frequently conducted . This can be achieved by using large language models together with code interpretation softwares like Python.
AI is the way to go but needs to be taught on how to use it
@Silungile, you've named the thing everyone else is dancing around. All the use cases in this thread assume people already know how to use these tools well and most of us are self-taught. What would "taught properly" look like at APHRC? A formal training, or something more like peer learning where people show each other what's actually working?
AI at APHRC is not just about
productivity, it is about paradigm. The biggest impact is not automating what
we already do, but redefining how evidence shapes policy in Africa.
At Purpose Rwanda, we are
exploring AI for predictive analysis of adolescent risk trajectories - linking
mental health, substance use, and school dropout to generate early,
rights-based interventions. Not to replace human judgment, but to extend it.
Two non-negotiables as we
advance:
1. Data sovereignty. African data
must serve African futures.
2. Ethical governance before
scale. Without African contextual training and human rights safeguards, we risk
reproducing harm.
The question is not whether we
use AI, but whether we govern it.
APHRC should lead that standard.
AI can strengthen our policy engagement by quickly analyzing large volumes of research data and distilling it into clear, actionable insights for government stakeholders. This would help us translate evidence into policy messages more efficiently.
I’ve used Microsoft Copilot to draft and refine internal reports and meeting summaries. It saved time by improving clarity and ensuring the language was policy‑friendly.But again I’d like to pilot AI‑driven policy briefs that combine APHRC’s research findings with real‑time government priorities, so our advocacy materials are both evidence‑based and immediately relevant to decision‑make
@Shamim Kowa Talla, your point about combining APHRC's findings with real-time government priorities is what I find most interesting, and the most dangerous. Real time priorities shift fast, and an AI tool optimised for relevance could end up telling us what a ministry wants to hear rather than what the evidence says. Have you thought about how you'd guard against that in a pilot? I'd be keen to see what a brief looks like when the evidence and the political moment pull in different directions.
@Bivon Momanyi, thanks for opening this up, the range of responses already shows how differently people are using "AI at APHRC." Where I think the biggest impact sits is not in any single task but in the connective tissue between them, taking evidence that already exists in separate projects and surfacing where it converges, so we stop commissioning new studies to answer questions our own data already answers. That is a coordination problem as much as a technical one. On tools, I've used AI mainly for qualitative coding support, first-pass thematic sorting of interview transcripts, with every code verified manually. It cut the mechanical work substantially, but the judgement stayed human.
The use case I'd like to explore is AI assisted peer review of our own outputs before they leave the building through checking draft reports and briefs for internal inconsistency, unsupported claims, and gaps between what the data shows and what the summary asserts. Essentially an adversarial reader, not a writing assistant.
Curious whether others see that as useful or as one more layer of process.