From Malawian farmers getting chatbot advice on crop diseases to Indonesian fishermen improving catches and Kenyan health workers diagnosing TB faster, AI is already being deployed across LMICs.
The most interesting question isn't whether AI delivers services more efficiently. It likely already does, and increasingly so. The interesting question is whether AI breaks the traditional development pathway entirely.
Historically, countries climbed the income ladder through a fairly predictable sequence:
Agricultural productivity gains
Then manufacturing
Then services
Lower labour costs plus strategic industrial policy helped Bangladesh dominate textiles and Vietnam become an electronics hub, lifting tens of millions from poverty. But that model depends on human labour costs mattering. If AI and automation make them less relevant, production could shift back to richer countries, and the ladder that built modern Asia might get pulled up.
Alternatively, if AI delivers personalised education, healthcare and cheaper energy at scale, we might see unprecedented acceleration in human development and massive reductions in the cost of goods and services. Much like WhatsApp unexpectedly becoming a foundation of commerce in LMICs - AI’s impact could be similarly unpredictable.
There could even be a decoupling of GDP from welfare with poor countries able to attain a much higher standard of living without increasing GDP.
I think both dynamics are likely to play out simultaneously, which makes this harder to reason about than most technology and development questions. What follows is my attempt to map the terrain. The real applications already working, economic scenarios for what comes next and what it might mean for LMICs.
Real World AI Applications: Transforming Development
AI is not a future concept in LMICs, it's a present reality, already being deployed to address pressing development challenges. These examples demonstrate AI's tangible impact across sectors, offering glimpses into how it can drive efficiency and improve access to goods and services.
Healthcare
AI is already a powerful diagnostic and research tool in healthcare, making higher quality services more accessible and efficient.
Computational Vaccine Design (AlphaFold & others)
Using AI to virtually design and optimise vaccine structures
A computationally designed COVID-19 vaccine induced approximately three times more neutralising antibodies than the Oxford/AstraZeneca vaccine, even at smaller doses
Radically reduces research and development time and cost, vital for rapid responses to health crises
Smartphones that analyse cough sounds (acoustic biomarkers) to screen for tuberculosis
Offers rapid, low cost screening without specialised equipment or extensive training, achieving up to 95% accuracy in clinical validations. This significantly improves the identification of presumptive TB cases and expedites referral for confirmation and treatment
Makes early, accessible diagnosis possible in remote areas, preventing further transmission and disease progression where diagnostic infrastructure is lacking
An AI enabled SMS service providing personalised pregnancy advice to mothers, processing 10,000+ daily questions and triaging high risk cases
Reached 2.74 million mothers across 1,200+ health centres in Kenya. Mothers were 22% more likely to achieve recommended antenatal care visits and 3.5 times more likely to seek care for danger signs, contributing to a 27% reduction in neonatal deaths at partner facilities
A scalable solution leveraging widespread mobile phone penetration to deliver health information and improve maternal and child health outcomes1
GiveDirectly's AI for Targeted Aid
Using AI to analyse mobile phone data for more accurate targeting of cash transfers for humanitarian relief
In Togo, this approach reduced the exclusion of eligible individuals by 8-14% for COVID-19 relief, ensuring aid reached those in poverty
Enhances the efficiency and accuracy of humanitarian aid and social protection programmes, ensuring limited resources reach the most vulnerable populations with less waste
Agriculture
AI is changing farming by providing above average insights, automating tasks and expanding access to resources for smallholder farmers.
Farmer.CHAT by Digital Green
Combines farmer-to-farmer videos with AI tools to deliver personalised agricultural advice through extension agents
Used by over 8.2 million farmers across India, Kenya and Ethiopia. RCTs show the approach is 10 times more cost-effective than usual services, reducing cost per adoption from $35 to $3.50, reaching 30% more farmers, and delivering a 43% gain in practice adoption rates
Democratises access to expert agricultural knowledge, allowing smallholder farmers to increase yields and income efficiently, a critical step for food security and poverty reduction
Drones configured to scatter seed, spray pesticide or spread fertiliser, operated via mobile phone app or AI automated
Cuts the requirements for managing some aspects of crop production by at least half. For one Vietnamese farmer, what took five days manually to spray a 30-hectare farm now takes four hours, using 30% less chemicals
Significantly boosts agricultural productivity, reduces labour needs and minimises chemical use, making farming more sustainable and profitable, even for smaller operations
Apollo Agriculture (Credit for smallholder farmers)
Uses machine learning and alternative data (like GPS) to provide loans, high quality inputs, agronomic training and insurance
Helped nearly 400,000 farmers in Kenya and Zambia achieve 2-2.5 times higher yields, boosting incomes and food security without requiring collateral
Manufacturing and Logistics
AI is optimising complex operations and cutting costs in critical industrial sectors.
Mathematical Optimisation for Cargo Ships
AI powered API for optimising shipping network design
Able to double the profit of a container shipper, deliver 13% more containers with 15% fewer vessels
Improves the efficiency of global supply chains, lowering trade costs, which benefits economies reliant on exports and imports
100 driverless electric mining trucks that autonomously load and unload materials in harsh conditions
Projected to improve transport efficiency by 20% compared to manned trucks, contributing to an 8% reduction in operational costs across China's coal mining sector
Jetstream Africa (Trade Finance)
Uses AI enabled tools, including optical character recognition, to streamline assessing credit risk and providing trade finance for businesses
Reduced credit decisioning time from an industry standard of around one month to less than one minute, resulting in loans and financial guarantees to hundreds of businesses in Nigeria and Ghana with a loss rate less than half the regional industry average
Education
AI is changing learning by personalising content and enhancing teacher effectiveness.
Rocket Learning (Early Childhood Learning, India)
An AI coach via WhatsApp supporting parents and childcare workers in low income communities by creating localised content and automating gradings
Reached 2.5 million children, with 75% hitting developmental milestones versus a 57% national average
Provides scalable, accessible early childhood education support, critical for foundational learning in contexts where formal schooling is limited or stretched
English Language Learning (Nigeria)
An AI powered tutoring programme for English language learning
An RCT found a significant overall improvement of 0.23 standard deviations in English language learning over six weeks, equivalent to 1.5 to 2 years of schooling. This positions it among the most cost-effective educational programmes
Science
AI is accelerating scientific discovery and forecasting critical environmental events, offering new tools for research and disaster preparedness.
Protein Structure Prediction (AlphaFold)
An open access AI model that predicts protein structures
Provides ~214 million predicted protein structures utilised by 1.6 million researchers across more than 190 countries, significantly speeding up research in areas from drug discovery to fundamental biology
Flood Forecasting (Google AI)
AI model predicting floods 7 days ahead
Covers 100 countries and 700 million people, sending over 115 million alerts in 2021 across India and Bangladesh. The World Bank estimates upgrading flood early warning systems could save 23,000 lives annually
Accurately predicted 70% of earthquakes a week before they happened during a seven month trial period in China
Offers the potential for disaster preparedness in earthquake prone regions, allowing for evacuation and mitigation efforts that could save lives and reduce economic damage. Earthquakes have been the deadliest natural disaster in the last few decades given how hard they have been to forecast
As these examples show, AI is already delivering results2 and improving lives across LMICs. However, the true breadth and depth of AI's economic implications, and whether it will fundamentally reshape development pathways, remains a subject of debate.
The Future of Economic Growth
The economic implications of AI sit at the heart of a central debate in modern economics. While there's broad consensus that AI will have meaningful effects, economists are divided on the magnitude, distribution and timeline of these changes, ranging from modest productivity gains to dramatic economic transformation.
Three Scenarios for AI's Economic Impact
This wide spectrum of uncertainty manifests in varying projections for AI's productivity impact, which typically cluster into three scenarios:
Conservative
This outlook anticipates AI will bring incremental, but still valuable, improvements
OECD projects 0.25-0.6 percentage points in annual total factor productivity growth3
Daron Acemoglu predicts a modest increase in GDP between 1.1 to 1.6% over the next 10 years (or ~0.1% increase a year)
Moderate
This perspective foresees substantial, but still manageable, boosts to productivity and economic output
Goldman Sachs envisions labour productivity (output per worker) boosts of up to 1.5 percentage points and a 7% increase in GDP over 10 years
Wiseman and McClements predict an additional annual economic growth boost of 3-9% in the near future
How much economic growth from AI should we expect, how soon?
Transformative
This scenario suggests that AI could become self improving, capable of automating knowledge production itself
AI transcends its role as a mere tool, evolving into an economic resource that compounds growth exponentially, potentially leading to explosive, unprecedented economic acceleration
Tom Davidson suggests there is a 1 in 10 chance of 30% annual growth rates by the end of the century4
Epoch AI - Erdil and Besiroglu estimate even odds of explosive growth by 2100
For those interested in exploring this in greater depth, BlueDot Impact offers a free course focused on the economics of transformative AI.
A Modern Solow Paradox
Yet, despite optimistic projections for AI's transformative potential, a confusing reality persists. Aggregate productivity growth remains low. This echoes Robert Solow's 1987 observation regarding computers - “You can see the computer age everywhere but in the productivity statistics.” Several explanations have been proposed for this:
Measurement Problems
Traditional economic metrics might struggle to capture AI's true value, missing improvements in service quality or the consumer surplus from free digital services
The value of personalisation or enhanced user experiences, central to many AI applications, frequently doesn't appear in GDP statistics
Implementation Lags
Historical patterns show that general purpose technologies, like the steam engine, electricity or computers, take decades to generate measurable productivity impacts
These technologies require significant complementary investments in skills and organisational restructuring, implying AI's full economic benefit is yet to materialise
Complex Propagation
Productivity gains from AI may propagate indirectly through intricate supply chains, appearing in downstream sectors rather than at the immediate point of AI adoption
A manufacturing firm using AI for logistics, for instance, might not show direct productivity gains itself, but enables efficiency improvements throughout its broader network
Threshold Effects
AI's systemic impact might only become visible once it reaches a critical mass across interconnected systems and industries, suggesting we are still in the early stages of economic transformation awaiting a tipping point for broad, measurable productivity acceleration
Distribution of Benefits
Beyond economic growth projections, a question concerns how AI's economic benefits could be shared. Economists hold differing views on whether AI will broaden prosperity or concentrate gains.
The Distributive View: AI as a Catalyst for Broad Prosperity
Rooted in historical precedent, this view sees AI as a general purpose technology that will broaden prosperity. Market mechanisms and falling costs are expected to distribute benefits widely, leading to a new era of abundance.
Augmentation Over Replacement
AI will augment human capabilities, fostering new, human centric industries and roles. Work will evolve, shifting towards uniquely human cognition, creativity and interpersonal skills
Market Driven Diffusion
As AI becomes cheaper and more ubiquitous, its benefits, like improved service access and reduced production costs, will diffuse through the economy, benefiting a broad consumer base. This could lead to a significantly higher standard of living due to drastically reduced costs of goods and services
Human Economic Value Endures
A core economic argument against extreme wealth concentration is the role of broad human consumption, without it, the economic pie could shrink. This implies an incentive for wealth distribution to maintain demand
The ‘Normal Technology’ View: Benefits Not Guaranteed
This perspective also positions AI as a general purpose technology, but one whose benefits are not automatically guaranteed to be widely distributed. It suggests that AI's integration will be gradual, and its impact will depend heavily on institutional responses.
Controllable Tool, Gradual Impact
Transformative economic and societal impacts will unfold slowly, over decades, bounded by organisational and institutional adaptation
This innovation diffusion lag stems from regulatory speed limits in safety critical domains, the ‘capability-reliability gap’ between benchmarks and real world utility, and the challenge of incorporating tacit knowledge. Benchmarks often measure methods progress, not actual utility
Systemic Risks
The primary concern is systemic risks common to large, interconnected systems (governments, large firms, oligarchies) which can be amplified by AI
These include exacerbated inequality, concentration of power and erosion of social trust
These arise from organisational choices, mirroring disruptions from past technology revolutions
Specific occupational displacement (copywriters, translators, etc) is a known risk
Institutional Capacity for Mitigation
Optimism lies in institutional capacity to respond
Policy can mitigate risks and distribute benefits through the usual regulations, investment in resilience and proactive redistribution
The Intelligence Curse: A Concentrated Resource
Another view suggests AI might behave less like a plough or steam engine (which augmented people and created new human centric industries) and more like coal or oil, a concentrated resource that can be more easily used to yield rent to its owners rather than broad prosperity
Diminished Incentive for Human Capital
If AI can reliably and cheaply perform most economically valuable work, powerful actors (states, corporations) may lose their incentives to invest in human capital
In a world where revenues flow primarily from AI systems rather than human productivity, the conventional drivers for widespread education, employment, and social safety nets (citizens' ability to generate taxes or profits, or their power to pose credible threats to a regime) could diminish
Resource Curse Analogy
This scenario mirrors the ‘resource curse’ seen in rentier states that derive wealth from natural resources rather than a diversified, human centric economy, often leading to concentrated wealth for a few and poverty for the general populace
If AI becomes the ultimate resource, its benefits could overwhelmingly flow to capital owners and those who control AI companies, displacing middle income employment without creating equivalent alternatives, leading to growth that does not translate into widespread prosperity
Challenging this Outlook
Governance Quality
The resource curse has been more prevalent in countries with poor institutions and high corruption
Nations like Norway and Australia have successfully leveraged resource wealth for broader benefit, suggesting that robust governance quality could mediate AI's distributional impact
Consumer Demand
If the general populace loses economic power, who will consume AI produced goods and services? A few wealthy individuals cannot sustain a large scale economy, potentially limiting profitability for AI owning entities
Lower Costs of Living
Additional mitigating factors include the potential for dramatically lower costs of providing a high quality of life with advanced AI, even with less traditional employment
Social & Political Agency
Society might designate certain ‘nostalgic’ jobs (priests, judges, etc) as exclusively human to preserve labour scarcity
Political action in response to job displacement could force redistributive policies to maintain social stability and avoid unrest
Implications for Development
The historical development trajectory, where countries typically progress from agriculture through manufacturing to services, faces disruption from AI. This transformation is simultaneously reshaping global trade patterns in ways that challenge conventional economic theories.
Technology Adoption
AI presents the potential for LMICs to leapfrog development stages by adopting AI enabled services (advanced diagnostics, automated governance) without first establishing conventional industrial infrastructure
The success of mobile banking and satellite internet shows this is possible
Effective adoption and widespread benefit rely upon infrastructure - reliable internet connectivity, affordable data and stable electricity
Many LMICs currently lack these elements, which risks creating a new 'digital divide' both within and between countries, marginalising rural areas or poorer populations who cannot access these services
This could slow down development but also insulate against potential job losses
Export Strategies
Automation poses a threat to the export oriented models prevalent in many countries heavily reliant on routine manufacturing
Bangladesh's textile industry or Vietnam's electronics assembly will see their competitive advantage erode as AI powered automation makes nearshoring or reshoring economically attractive for richer nations
This may compel these countries to focus on higher value niche services, creative industries, or human centric labour
Fostering economic resilience may involve prioritising the lowering of intra-regional trade barriers, for example, within Africa and Asia, and forming strategic partnerships, rather than relying solely on historical global market access which may be shifting
Digital Services
AI enables the emergence of new categories of tradeable digital services (data labelling/annotation, content moderation and other sectors that require human in the loop roles like medicine or law) offering a potential pathway that bypasses industrial development
A country could conceivably export AI generated content, automated customer service or data processing services without needing to establish a conventional manufacturing base
Concentration Effects
The development and deployment of advanced AI depend on computational resources, highly specialised talent pools and access to data
These advantages are concentrated in the richest economies, fostering winner takes all dynamics
This risks increasing global inequality, potentially locking countries out of lucrative and transformative sectors of the future economy and amplifying existing disparities
LMICs could benefit from forming coalitions to negotiate more effectively with global AI developers to ensure that technology supports national development agendas
Domestic Labour Markets
Beyond export oriented jobs, AI could impact domestic labour markets within LMICs, affecting service sectors, agriculture and the informal economy
Widespread job displacement in routine tasks could lead to increased internal inequality, particularly between those able to adapt and those unable to do so
Human Capital Development
The skills demanded in an AI driven economy are likely different from those normally required in manufacturing. LMICs face a challenge in retraining existing workforces and fundamentally reorienting their education systems
A failure to address this skills gap could inhibit the 'leapfrogging' potential AI offers
Governance
AI offers tools for improved governance, such as enhanced public service delivery, more efficient resource management or crime prevention
It also could be used to help entrench authoritarian regimes, and make bad regulations easier to monitor even if they are damaging to an economy
Broader Societal and Ethical Risks
Models are often predominantly trained on data from a few high income countries, reflecting their specific social categories and labour market structures
When deployed internationally, such systems risk importing these biases, which may interact unpredictably with local social hierarchies and norms which could lead to a homogenisation of hiring practices globally or less useful/more harmful applications
Environment
While the direct energy use of AI queries is low for individuals, the overall electricity consumption of data centres for AI is a substantial and increasing concern
Intensive mining of minerals like copper and lithium can cause ecosystem damage, water depletion and pollution
At the same time, AI can be used to reduce energy costs and accelerate research into solar, battery, nuclear technologies, etc
AI models and associated applications can be very different month to month, posing a challenge for academic research cycles to keep pace. To remain relevant, development research may need to focus more on identifying underlying mechanisms and principles that outlast any single AI model or interface or for academia to significantly speed up research cycles.
AI Strategies for LMICs
Given these implications, the question for LMICs becomes how to strategically engage with AI and related policies to benefit their countries which may trade off with the risk of increased harms.
Infrastructure and Connectivity
Internet access reaches only 27% in low income countries versus 93% in high income nations, with broadband costs accounting for 31% of monthly income in low income countries compared to just 1% in wealthy ones. Some countries have regular power outages and internet failures/blackouts
These constraints suggest developing tools that rely on less energy/internet access or focusing more on building up energy infrastructure and utilising satellite internet
Foundational AI vs Applications
Some people suggest that countries should develop their own foundational models to have more control over their future
But the resource gap is huge. The US secured $67.2 billion in AI investments in 2023 and is one of few countries that could develop frontier models
Building foundational models from scratch costs tens if not hundreds of millions of dollars and requires a lot of pre existing infrastructure whilst adapting existing models is far less costly
The choice mirrors LMIC startup strategies. Just as companies like Jumia became ‘Africa's Amazon’ without reinventing ecommerce infrastructure, AI applications can solve local problems using existing models. Technologies like mobile based commerce and banking have been adopted faster in LMICs compared to high income countries, supporting the idea that countries can leapfrog in AI adoption with the right conditions
Avoiding Hype
Governments may announce ‘AI transformation’ programmes but no clear theory of change linking AI deployment to improved outcomes. These initiatives often exist primarily for political signalling rather than addressing concrete policy problems
Funders may mandate AI quotas (‘allocate 15% of your programme budget to AI solutions’) creating incentives to retrofit AI onto problems that may be better solved in other ways
Talent and Brain Drain Challenges
The global talent distribution is heavily skewed, nearly 60% of all top tier AI researchers reside in the US, six times the number in China and Europe, whilst India has ~400 people out of the 22,000 PhD educated AI scientists globally
This reinforces the application focused strategy. Rather than competing for scarce foundational AI talent, countries can build practical expertise in adapting and deploying AI solutions. The skills needed for effective AI application development are more about understanding local contexts and navigating regulatory environments
Regulatory Environment
Major tech companies including Google, Apple and Meta have delayed AI product launches in Europe due to regulatory uncertainty, whilst offering full features in less regulated markets
The EU's AI Act creates compliance burdens that extend time to market and give competitive advantage for providers operating in jurisdictions with lower regulatory standards
Unlike the extractive compliance typical in LMICs (where regulatory systems often serve elite interests rather than development goals), AI regulation presents a different dynamic that may not be captured by elites yet
Success stories like Ghana's minohealth AI Labs, which developed radiological diagnosis systems now used globally, demonstrate how LMICs could move faster than Europe to deploy and export AI solutions. This represents a genuine first mover opportunity for practical applications
Data Quality
LMICs often rely on models developed by large tech companies with training data that might be less useful in their country, whilst facing challenges like a lack of access or even a lack of data in the first place
These limitations create opportunities as well. Local applications can address specific cultural and linguistic needs that global models might miss. The key is building solutions that work with available data whilst gradually improving local data collection capabilities
Career Pathways in Emerging Technology
For people motivated to contribute to ensuring AI benefits global development and LMICs, there are various pathways available across technical, policy and implementation.
Frontier AI Companies
Working at major AI companies to influence how foundational AI systems are developed and deployed globally.
Paths to Impact
Help ensure that foundational models work effectively in multiple languages and can be deployed in low resource environments
Work on making AI models more efficient so they can run on basic hardware and limited internet connectivity
Influence product development to consider applications that could dramatically improve healthcare, education or agricultural productivity
Example Roles
Research scientist roles focusing on model efficiency or multilingual capabilities, product manager positions for global AI deployments, technical roles working on reducing computational requirements, business development roles identifying beneficial use cases in emerging markets
Pros
Working with the most advanced AI systems and largest budgets for R&D. If you can influence foundational models to work better, the scale of impact could be enormous since these models underpin many applications
Cons
Your influence on company priorities is likely to be quite limited unless you reach senior levels. Most frontier AI companies are primarily focused on high value markets in wealthy countries
AI Startups & Businesses
Building commercially viable AI solutions in sectors like agriculture, healthcare, education or financial services for LMIC markets.
Paths to Impact
Create businesses around AI applications that help farmers increase yields, enable better healthcare diagnosis or improve educational outcomes
Develop approaches to reach populations that most tech companies don't serve
Build solutions that work offline or with limited infrastructure
Example Roles
Founding teams focused on agricultural technology, health tech, or fintech, technical roles building AI tools that work on basic smartphones, business development roles expanding AI solutions to new markets, product roles designing for low resource environments
Pros
Market incentives align with user needs - if your product doesn't actually help people, they won't pay for it
You can iterate quickly and get direct feedback
Commercial sustainability means you're not dependent on donor funding cycles
Cons
The people who most need help are often the least able to pay for it, creating a fundamental tension
Many AI applications require upfront investment with uncertain returns.
Operating in LMICs often involves infrastructure challenges, regulatory uncertainty and currency risks that make businesses harder to scale
Government & Policy
Shaping how AI is regulated and deployed to maximise benefits for economic development.
Paths to Impact
Help governments create policies that enable beneficial AI applications whilst avoiding regulatory barriers that prevent innovation
Work on strategies that help countries leapfrog development stages using AI
Support efforts to ensure AI deployment increases rather than decreases economic opportunities
Example Roles
Policy advisor roles in country governments working on digital strategy, positions at international organisations like the World Bank or USAID working on AI and development, roles helping navigate AI regulation across different countries, positions working on trade policy that affects AI deployment
Pros
Governments have significant influence over whether beneficial AI applications can actually be deployed at scale
Policy work can have large multiplier effects if you get the frameworks right
There's relatively little expertise in this intersection, so individual contributions may matter more
Cons
Government decision making is often slow and influenced by political rather than evidence based considerations
Your impact depends heavily on political stability and whether the people you're advising remain in power
Many governments have limited implementation capacity even when they have good policies
Academia, Think Tanks & Nonprofits
Conducting research, generating evidence and implementing AI solutions in development contexts.
Paths to Impact
Generate evidence on which AI interventions improve people's lives and livelihoods
Bridge the gap between what's technically possible and what works in practice
Train people to work effectively at the intersection of AI and development
Example Roles
Researchers studying AI's impact on poverty reduction, field implementation roles with organisations deploying educational technology or agricultural advice systems, positions at think tanks studying AI's economic effects, programme roles at foundations funding AI applications in health, education, or agriculture
Pros
You can work on neglected questions that other sectors won't fund
Academic and nonprofit environments can allow for longer term thinking and risk taking
Direct connection to outcomes and evidence of what works
Cons
Limited resources mean you often can't implement solutions at the scale needed to significantly impact poverty
Publication incentives in academia may not align with practical impact
Many pilot projects fail to achieve sustainable scale
Grant funding is competitive and often project based rather than allowing for sustained work
Key Considerations
The intersection of AI and development offers opportunities, but success requires understanding both technical capabilities and practical constraints.
The field is evolving rapidly, and many challenges are not necessarily AI specific, they're about deployment, adoption and fitting into existing systems.
Evidence on what works is still limited, and there's significant risk that AI applications may not deliver the promised benefits or could even create new problems. However, the potential upside is substantial if AI can help accelerate development.5
Further Resources
Dylan Matthews - How AI could explode the economy
Frontier Tech Hub - AI as a tool for International Development professionals
SSIR
How AI powered nonprofits could make health care more effective
Mapping the Landscape of AI Powered Nonprofits
Alice Evans - Crafting AI Complementary Skills and Bulletproof Assessments (at universities)
Google
Health AI Developer Foundations is a new suite of open weight models to help developers more easily build AI models for healthcare applications
1000+ real world AI use cases from organisations
Peter Breitbart - AI for Doing Good: Lessons from the Frontlines
VoxDev - collection of AI articles
Nature - Can AI help beat poverty? - Measuring poverty is the first step to delivering support, but it has long been a costly, time intensive and contentious endeavour
Turn.io looking at how chat and AI has been used to achieve impact in 2024
That’s not to say that all AI projects work, here is a case study from GiveDirectly where they delivered digital cash aid to flood survivors in Nigeria and Mozambique. However, in Mozambique, challenges arose with the AI driven flood forecasting models which lacked the precise accuracy to predict severe flooding at the specific village level, leading to anticipatory payments in areas that ultimately weren't severely hit
A broad measure of efficiency in how inputs are used
Although this is a different time frame to the conservative estimates above and both predictions could theoretically exist in a world with later take off timelines
Questions for reflection
How does AI compare to previous transformative technologies in terms of development potential?
How likely do you consider the various forecasts for AI driven economic growth?
Despite rapid advances in AI, aggregate productivity growth remains low, why do you think this is?
Beyond economic metrics, what fundamental shifts in human experience and societal structures could AI bring to LMICs that differ from previous technological revolutions?
Applications and Impact
Which AI applications seem most promising for improving lives?
How should we prioritise AI interventions versus proven non AI approaches?
What development challenges seem more suited to AI solutions?
Beyond specific applications, what mechanisms or enabling conditions are most needed for positive impact or technological leapfrogging?
How can the tension between building commercially sustainable solutions and serving the needs of the poorest populations be overcome?
If AI can genuinely offer much better education, healthcare, etc is it ethically justifiable not to prioritise rapid, widespread deployment, even with uncertainties?
Should AI be used to automate governance itself?
Risks
What are the biggest risks of AI adoption?
How might AI adoption impact labour markets and employment patterns?
If AI drives massive productivity gains but concentrates wealth and power, are countries facing an ‘intelligence curse’ worse than the ‘resource curse’ of the past?
What risks of AI adoption do you think are underestimated or overlooked by the development community?
If LMICs choose to leverage less stringent AI regulation as a first mover advantage, are they playing a risky game that could lead to unforeseen harms, or seizing a vital opportunity to shape their own technological destiny?
Future Scenarios
Who should set the ethical and regulatory standards for AI in development contexts - local communities, national governments, international bodies or tech companies?
What specific governance approaches are needed for AI in development?
How should the development ecosystem adapt to an AI enabled future?
Which non AI technologies deserve more attention from the development community?
Careers and Personal Impact
What skills should development professionals prioritise learning about AI?
How can individuals contribute to beneficial AI deployment in LMICs?
Which sectors might offer the highest leverage to impact the future of AI?
Top AI companies
Startups and business utilising AI
Governance of AI
Academic research
Really thoughtful and well-balanced piece. There’s no doubt AI will shape the future of developing economies — but I’m a bit skeptical that it will actually close the global gap. The original Great Divergence began when large enterprises emerged in the West, employing most of the workforce. In 19th-century Britain, for example, the shift from self-employment to wage labor was massive. At the same time, accounting and auditing practices evolved to assess investment risk and potential — fueling enterprise growth. (I wrote more about this here: https://exchangism.substack.com/p/exchangism-how-market-expansion-has?open=false#%C2%A7iii-large-enterprise-made-possible-by-global-market
)
Without a similar expansion of business institutions that can absorb and organize labor, I fear the divergence will persist — even in the age of AI.