Putting AI in aid: new tools for development policy and programming
The integration of artificial intelligence into international development is beginning to move from speculative discussion to practical application. At the 2024 Australasian AID Conference, one session explored how AI tools are being developed and tested to enhance development policy and programming, with potentially interesting implications for how aid organisations analyse problems and make decisions.
Two organisations with an interest in this field – Dragonfly Thinking and the Development Intelligence Lab – demonstrated how structured AI applications could augment “analog” (that is, human) analysis while avoiding some of the pitfalls commonly associated with AI adoption.
Miranda Forsyth, co-founder of Dragonfly Thinking and a professor at ANU’s School of Regulation and Global Governance (RegNet), described her journey from AI sceptic to innovator. Despite being “one of the most non-technological people around,” Forsyth’s collaboration with Professor Anthea Roberts led to the development of an AI platform that helps users think through complex development challenges.
At the heart of their approach is the “Triple R Framework” – based on an examination of risks, rewards, and resilience. Unlike binary or pro-versus-con approaches such as SWOT or cost-benefit analysis, this framework adds a third dimension, resilience, that takes into account how systems change over time. Risks arise at the intersection of threats, vulnerabilities and exposure, while rewards arise at the intersection of opportunities, capabilities and access. The resilience element involves three distinct capacities: absorptive (for immediate responses), adaptive (for real-time strategy adjustments), and transformative (for system-level changes).
What distinguishes Dragonfly Thinking from conventional AI tools is its cognitive architecture, which is layered over large language models like ChatGPT. Rather than simply generating responses, the system guides users through a process of structured analysis and incorporates multiple perspectives – much like its namesake insect’s compound eyes, which provide 360-degree vision.
Forsyth demonstrated the platform’s application to a consideration of sorcery accusation-related violence in Papua New Guinea, showing how it could help analyse social media’s role in either driving or preventing violence. The system helps users think through immediate threats (like viral videos of violence), exposure factors and vulnerabilities, as well as opportunities for human rights defenders to counter harmful narratives.
The Canberra-based Development Intelligence Lab has been experimenting with this technology over the past year, testing whether combining “analog analysts” (again, those are people) with AI can produce better policy analysis. CEO Bridi Rice shared results from an experiment where analysts with no particular expertise in development economics, under time pressure, were tasked with analysing middle-income traps in Southeast Asia.
Within just two hours, using the Dragonfly platform, the team produced an analysis that development economists validated as providing solid insights. Notably, the AI-assisted analysis offered a more comprehensive view than some human analysts might have, given their self-recognised tendency to focus predominantly on either risks or opportunities, depending on their disciplinary background.
Rice emphasised that the tool’s primary value lies not so much in the efficiency gains it offers but in its ability to help people “think better” and, perhaps, make better decisions. While acknowledging serious concerns about AI bias – particularly in relation to Western knowledge dominance and the limited incorporation of local perspectives – she argued that the technology shows promise in overcoming some of the disciplinary and institutional biases that can plague policy discussions.
The organisations’ experience offered some lessons for development organisations considering AI adoption. First, while AI can enhance efficiency in routine tasks such as document synthesis, its greatest potential might lie in improving the quality of analytical and decision-making processes. Second, while bias remains a significant concern, structured analytical frameworks can also help reveal and potentially overcome certain types of institutional and disciplinary biases.
It was noted that one of the more promising aspects of the technology could be its potential to enhance local voice and perspective in development decision-making – not least as a result of AI’s rapidly advancing translation capability. The Development Intelligence Lab’s “Southeast Asia Pulse Check” project, which aggregates expert insights from up to 50 people in each Southeast Asian country, demonstrates how AI tools might help bridge the gap between local knowledge and policy formulation. This could be particularly valuable for smaller nations that lack extensive research and policy analysis capability.
The platform’s ability to analyse issues from multiple perspectives could also help address a common criticism of development programming – that it often reflects donor preconceptions rather than local realities. By systematically incorporating different viewpoints and knowledge systems, AI-assisted analysis could help development practitioners better understand and respond to local contexts.
Both organisations announced initiatives for 2025, including an “AI in development” discussion group open to practitioners and researchers, and opportunities for other organisations to partner with them on using the technology for specific policy challenges. They’re also working to incorporate more diverse knowledge sources and perspectives into the system, including exploring ways to better capture and integrate oral knowledge and indigenous perspectives.
However, some difficult questions remain about AI’s role in development. These include managing its intensive energy requirements – with AI searches consuming approximately ten times the energy of standard internet searches – and ensuring appropriate incorporation of indigenous and local knowledge. Both presenters were at pains to emphasise that AI should complement rather than replace deep consultation and local engagement in development program design and implementation.
A further consideration is how to maintain appropriate human oversight as these tools become more sophisticated. As Rice noted, while some senior policy makers are already impressed with the quality of AI-assisted analysis, the goal is to enhance rather than replace human judgment in decision-making. AI is an assistive technology, not a replacement for human experience and judgement.
The development community — like pretty much all other non-corporate users — will also need to grapple with questions of access and equity in AI adoption. As tools become more sophisticated and valuable for policy analysis and program design, ensuring equitable access across different development organisations and contexts will be crucial. This includes considering how smaller organisations and those in developing countries can benefit from these technologies without being disadvantaged by resource constraints.
So far, the collaboration between Dragonfly Thinking and the Development Intelligence Lab suggests that thoughtful AI adoption, focused on augmenting rather than replacing human capabilities, has the potential to help development practitioners navigate increasingly complex challenges more effectively.
This article was originally published on Dev Policy Blog by the Development Policy Centre, ANU Crawford School of Public Policy, and is licensed under a Creative Commons Attribution Noncommercial Sharealike 4.0 International License.