Small AI, Big Impact: Reflections from the IMF - World Bank Annual Meetings in Washington DC

By Rohit Kakanoor, Global Voices Fellow, IMF + World Bank 2025 Annual Meetings

Artificial intelligence came up constantly during the IMF - World Bank Annual Meetings in Washington DC, but not in the dramatic, futuristic way it usually appears in the media. Instead of discussions about frontier models or existential risks, many conversations focused on something far more grounded: Small AI. This idea centres on lightweight, low-compute tools designed to solve practical problems in areas like public administration, health, agriculture and financial inclusion. What stood out to me throughout the week was how consistently speakers emphasised that countries don’t need the most advanced systems to see impact. They need technology that works within their infrastructure, budgets and institutional realities.

One of the strongest insights for me was how Small AI can genuinely strengthen government capacity. Many public institutions, especially in developing economies, operate with limited staff, paper-based processes, long backlogs and data that is often incomplete or inconsistent. Small AI tools are already helping automate simple administrative tasks, classify documents, manage citizen requests and reduce turnaround times for core services. None of this replaces public servants; instead, it frees them from repetitive tasks so they can focus on what requires human judgment. This practical, incremental approach was a refreshing contrast to the global narrative that often jumps straight to large, complex models. It showed how meaningful improvements to service delivery can come from very simple, well-designed tools.

Financial inclusion was another recurring theme. In many developing markets, millions of people still can’t open a bank account or apply for credit because they lack formal documentation or financial histories. Small AI can help bridge these gaps. Identity verification tools that work with low bandwidth allow onboarding even where connectivity is weak. Alternative-data credit models can evaluate risk for people who would otherwise be left out of the system. And simple digital advisory tools help small-business owners understand their cashflow and make better financial decisions. Having worked on digital onboarding and eIDV in Australia, I found it striking how similar the principles are, even though the environments are different. At its core, an inclusive financial system depends on technology that is clear, fair and designed around the needs of real people.

A consistent thread across the Meetings was the importance of responsible governance and local ownership. Many countries are cautious about adopting large, opaque systems that may not interpret their languages, cultural norms or regulatory requirements accurately. The value of Small AI here is that it can be adapted locally. Models can be built around local dialects, agricultural cycles, environmental patterns or policy rules without needing massive computational resources. There was a strong emphasis on developing domestic capability - building local data expertise, ethical oversight and technical teams - so countries can manage and evolve their systems independently rather than relying entirely on external providers.

What became increasingly clear is how different the development conversation is compared with broader global debates about AI. In high-income countries, AI is often framed around cutting-edge research, productivity disruption and national competitiveness. But for many emerging economies, the biggest gains come from smaller, targeted interventions. A lightweight tool that helps teachers generate lesson plans, a system that digitises patient notes in a rural clinic, or a model that sends tailored farming advice can have more immediate impact than any sophisticated frontier model. This perspective connects closely to my own work in Australia on improving access to homeownership. Very often, the best solutions are not the most complex—they are the ones that fit the context.

Reflecting on the Fellowship, several lessons stayed with me. First, many AI challenges are data challenges. Without reliable records - whether civil, land, health or financial - AI systems have limited room to operate effectively. Strengthening these foundations is essential. Second, capacity-building matters just as much as the technology itself. The long-term success of AI depends on investing in local analysts, policy teams, data stewards and technical specialists. Third, solutions must be co-designed with the people who will use them. Technology only works when it reflects local priorities, culture and practical needs. And finally, even the simplest AI tools can unintentionally widen inequality if they’re not designed carefully. If certain populations aren’t digitally connected or if historical bias seeps into the data, the benefits of AI can become uneven. Ensuring fairness and accessibility from the outset is critical.

Overall, the Annual Meetings offered a very grounded and hopeful view of AI’s role in development. The future of inclusive technology won’t be driven by the size or sophistication of models, but by how well they serve the people they’re built for. Small AI reflects a more democratic and pragmatic approach - one that strengthens institutions, broadens access and delivers real impact in everyday settings. The most transformative tools for many countries won’t come from frontier labs, but from simple, thoughtful systems that solve real problems in effective and human-centred ways.

-------

The views and opinions expressed by Global Voices Fellows do not necessarily reflect those of the organisation or its staff.