AI might improve global health, but it’s a double-edged sword for development

This is a summary of original research of Rethink Priorities, conducted by Jamie Elsey, Ruby Emerson, Kieran Greig, John Firth, and Jerome Mayaud.

The global health and development community is understandably excited by the possibility that AI might facilitate, or even supercharge, their impact. Examples such as AI-assisted diagnostics, faster drug discovery for neglected diseases, and smarter weather forecasting for smallholder farmers are tractable applications with the potential to genuinely reduce suffering. We, too, have written in the past about the promises of AI applications for global health and have considered how philanthropists could leverage them carefully.

But do such improvements represent a transformation of the fundamental drivers of long-run development? In a new Rethink Priorities report summarized below, we aimed to answer this question by engaging with different lenses through which development economics has understood the constraints facing low- and middle-income countries (LMICs).

AI shows real promise in addressing natural and geographic constraints, such as tropical diseases or poor soils:

• AI-assisted diagnostics can extend specialist expertise to under-resourced settings.

• Protein-folding prediction and computational compound screening could compress timelines for treatments of neglected tropical diseases (conditions that account for roughly 12% of the global disease burden yet represent under 1% of research and development (R&D) spending).

• AI-improved crop breeding and weather forecasting could address documented productivity gaps for tropical agriculture.

These are just a few examples where AI might unlock meaningful welfare improvements. But technology can improve welfare without shifting growth, which itself drives longer-term development.

Mobile money in East Africa (M-Pesa) is an illustrative case: it reduced poverty and improved resilience during shocks, but there is considerably weaker evidence that it drove general productivity growth.

Similarly, AI diagnostics still require delivery systems (infrastructure & human capital) to reach the end user, trained workers, and functioning supply chains. Drug discovery will likely continue to prioritize commercial disease markets unless deliberately redirected toward the disease burdens in low- and middle-income countries (LMICs). Agricultural applications depend on digital infrastructure and face the same adoption challenges that have hampered agricultural extension for decades. Financial inclusion applications similarly depend on digital financial infrastructure that many low-income countries still lack.

AI investments in health, agriculture, and financial inclusion are still worth pursuing. But they should be bundled with delivery-system investments to reach intended users, held to the usual evidence standards, and not treated as a higher priority solely because they are AI-driven.

The functioning of institutions has been recognized as a fundamental driver of long-run growth: inclusive institutions (those that broadly secure property rights and constrain elite power) create incentives for investment and innovation, whereas extractive ones concentrate power and redirect incentives toward rent-seeking. We believe the potential impact of AI on institutions is the most significant blind spot in the global health and development (GHD) discourse on AI’s implications.

• Where institutions are already reasonably functional, AI tools can help build state capacity through innovations such as automated tax administration, AI-enhanced health information systems, and satellite-based land monitoring. But these tools require political will and institutional strength to be used productively. Deploying AI without that foundation risks “isomorphic mimicry”: adopting the form of good governance without the underlying substance.

• A more alarming trajectory is one in which AI adoption actively worsens governance. AI-facilitated mass surveillance and information control lower the cost of suppressing dissent relative to that of delivering public goods, thereby decoupling regime survival from economic competence. This removes one of the central growth-forcing pressures on governments. Some authoritarian regimes have already shifted toward managing citizens through information control rather than overt coercion; AI could further this trend.

• Global AI and associated energy demand may also exacerbate ‘resource curses’ in LMICs: extracting natural resources has historically been a way in which authoritarian leaders have enriched themselves and their patrons without benefiting the wider citizenry. Even absent corruption, much of the value from such extraction often accrues to other countries where resources are refined and transformed into higher-value products. The geography of AI- and energy-critical minerals, including lithium, cobalt, copper, and rare earths, maps closely onto regions of institutional fragility. The Democratic Republic of Congo, for example, contributes considerably to global cobalt and copper supplies, but is one of the poorest countries.

• Finally, automation may erode the mechanism through which citizens have historically extracted better governance: where elites depended on productive labor, they had incentives to invest in human capital and infrastructure. Widespread domestic automation could weaken this bargaining chip if it fails to unlock new and important roles for workers (see below for risks of foreign automation).

The structural transformation tradition identifies labor-intensive export manufacturing as the most historically reliable route to income convergence between LMICs and high-income countries (HICs). AI-driven automation in HICs threatens to erode the low-wage labor advantage that enabled this path.

• If AI makes it cost-effective to automate production domestically, the incentive to offshore to lower-income countries diminishes, and the entry point to industrialization narrows or disappears.

• We are also skeptical that services can cleanly substitute manufacturing. The service-sector alternatives that some economists identified as a secondary development path, such as call center work, data entry, and basic coding, include precisely the tasks now most exposed to AI automation. Data labeling and similar ‘gig’ work have been documented as low-wage and precarious, offering little pathway to higher-value participation in the AI economy.

• Meanwhile, the value in AI supply chains is concentrated at the top: over 90% of frontier AI compute capacity is currently held by the US and China, and in AI research, the African continent accounts for under 1% of global output. Countries that are purely consumers of AI tools developed elsewhere may experience labor displacement without the corresponding economic returns.

The core strategic implication of our research is that GHD conversations about AI should focus on trade and foreign investment policy, not just on how AI might alleviate recognized burdens in areas such as health and agriculture.

This requires two distinct shifts in the GHD community:

• Support cross-disciplinary work that bridges GHD and AI policy, such as investigating whether differential access regimes could empower benevolent institutions and deter harmful applications of AI in developing countries, and how to balance state technological sovereignty with restrictions on nefarious uses of AI.

• Raise intellectual property/access regimes, LMIC regulatory capacity, and mineral supply chain governance on the development agenda to enable capability building and spillovers rather than dependency and extractive dynamics.

• Elevate trade and industrial policy to a central role in GHD discussions about AI, not just AI applications within existing or circumscribed interventions. Formal modeling of different scenarios of AI impacts on trade and the global economy may prove vital in informing such discussions

• Treat political will as a key investment criterion for AI governance solutions. Technology won’t compensate for the absence of political will, and it risks creating the illusion of good governance without its substance.

Even where we are cautiously optimistic about AI in health, agriculture, and financial inclusion, we think better outcomes will result when funders and GHD specialists:

• Bundle AI tools with delivery system investments: Co-invest in the local infrastructure and human capital that make tools usable.

• Maintain evidence standards: AI-driven interventions should not be prioritized on novelty alone.

The full report, What Do Increasing AI Capabilities Mean For Global Health and Development?, is available on Rethink Priorities’ website.

For any questions about this work, please reach out to ghd@rethinkpriorities.org.

Thank you for taking the time to read our Substack. We will continue to provide this content on a regular basis. If you would like to support our efforts, please subscribe below or share our posts with friends and colleagues.

We are also always looking for feedback on our work. You can share your thoughts about this publication anonymously or reply to this email/post with suggestions for improvement or any questions.