AI / Governance / Global Policy
Building the Capacity to Benefit from AI
Practical pathways for countries with different starting points
Income gap between the richest and poorest countries in Asia-Pacific (GNI per person)
Rural residents in Asia-Pacific who can use a basic spreadsheet formula
Executive Summary
The Industrial Revolution split the world into a few rich countries and many poor ones. AI is a technology of similar reach. It can tutor children in their own language and help short-staffed clinics read X-rays. But its gains could stay with the countries already ahead.
In December 2025, UNDP's Regional Bureau for Asia and the Pacific published The Next Great Divergence: Why AI May Widen Inequality Between Countries. The report asks who will gain from AI and who will bear its risks in the region with the widest income gaps between countries. It draws on nine background papers by researchers at institutions including MIT, LSE, the Max Planck Institute for Human Development, Tsinghua University, the University of Science and Technology of China, the Aapti Institute and Digital Futures Lab.
In a Brookings essay, Michael Muthukrishna and Philip Schellekens describe two gaps. The capability gap means AI's gains arrive first where skills, infrastructure and institutions are strongest. The vulnerability gap means its disruptions fall hardest on the people least prepared to manage them.
Why AI could split the world again
Two centuries ago, the world was poor almost everywhere, and no region had a life expectancy above 40. The Industrial Revolution changed that unevenly: by 1950, a Norwegian could expect to live 72 years and an Afghan 28. AI is a general-purpose technology like steam power and electricity, and it is spreading faster than any before it. ChatGPT reached a million users in five days. But as with electricity, which took half a century to transform factories, the report expects AI’s economic gains to come slowly, and to reach first the countries that can afford to reorganize around it.
AI is arriving in the region with the widest income gaps
Nearly 200 times apart
In 2024, Afghanistan's income per person (GNI per capita) was under $400 and Singapore's over $70,000. No other world region has a gap this wide. The report puts the gap at about 70 times after adjusting for prices.
Countries within countries
Kuala Lumpur and Shanghai both have GDP per person of about $30,000. Kelantan, Malaysia's poorest state, would rank in the lower half of Asia-Pacific countries by that measure.
A third of the region is offline
In 2024, 66% of people in the region used the internet. In 2019, 70% of urban households had access, almost double the 37% of rural households.
Fewer than 1 in 5 rural residents can use a basic spreadsheet formula
Among urban residents the share is about 1 in 4.
With gaps this wide, an AI strategy that works in Singapore would not work in Papua New Guinea.
How AI could widen gaps
The report traces AI's effects on people, the economy and governance, and finds both promise and risk in each.
People it can't see
Eligibility for public services, credit and social protection increasingly depends on data. People missing from the data risk being invisible to AI. Women, older people, rural communities, minorities and displaced people are the most likely to be missing. India's 2016–17 Economic Survey cited a 49% authentication failure rate in Jharkhand for Aadhaar, the national biometric ID, and noted that such failures shut genuine beneficiaries out.
Work it disrupts unevenly
About 1.3 billion people in the region work informally, often without social protection. Worldwide, women's jobs are nearly twice as likely as men's to be highly exposed to automation by generative AI. Poorer economies face less immediate automation, because more of their work is manual or offline. They are also less able to capture AI's productivity gains, and their safety nets are thinner.
Decisions that are hard to contest
Governments with less technical capacity can become dependent on foreign systems they cannot audit or adapt. When such a system helps decide who gets welfare, errors are hard to contest and responsibility blurs.
Much of the report's advice is about building registries that include everyone, clear rules and ways to appeal.
The principles behind the recommendations
Put people first
The report judges AI by what it does for people's freedoms and capabilities, not by how advanced it is. It asks for equity by design, meaning systems built from the start with marginalized groups in mind. It also wants teachers, health workers, rural communities and others most affected to help shape AI systems and to be able to challenge the decisions those systems make.
Govern innovation responsibly
Safeguards should scale with risk. A chatbot on a government website needs lighter rules than a system that decides who is eligible for surgery. High-stakes uses call for transparency, human oversight and a route to appeal.
Build future-ready systems
Future-ready systems need local talent and sustainable infrastructure. They also need open, competitive markets, so that no single vendor or country can lock others out.
In her foreword, UNDP Regional Director Kanni Wignaraja says the report's recommendations “should resonate globally,” because Asia and the Pacific holds more than half the world's population and because the questions the report raises apply wherever UNDP works.
Philip Schellekens
Chief Economist for Asia and the Pacific, UNDP
Roadmaps for different starting points
The report sorts its recommendations by principle, by a country's starting point, by time horizon (the first year, one to two years, three to five years) and by sector.
It advises each country to start with a candid readiness assessment of two kinds of capacity.
Neither works alone. “Hardware without human capacity wastes investment; ethics without access excludes by design,” the report says. It cites One Laptop per Child, which set out to give affordable laptops to children worldwide and struggled because it did not adapt enough to local circumstances. Students often lacked the skills and support to make full use of the devices.
First moves for three starting points
Lower capacity: Cambodia, Papua New Guinea, Lao PDR
The report gives priority to affordable connectivity, low-cost digital payments and a digital ID with strong consent and privacy protections. Alongside these foundations, AI that works offline can help health workers with triage, give farmers spoken advice in local languages and warn communities of hazards.
Transitional capacity: India, Indonesia, Viet Nam
Most people have internet access and ministries have run AI pilots, but coverage and quality remain uneven. To turn pilots into services people can rely on, the report calls for infrastructure that stays up, datasets that represent the whole population and support for workers whose jobs change. It also advises against depending on a single vendor.
Higher capacity: Singapore, Australia, Republic of Korea, China
These countries can lead on regional standards, safety and green infrastructure. They can also build what smaller neighbors cannot fund alone, such as regional computing capacity, shared AI models and datasets in the region's languages. The report asks them to transfer technology and knowledge to the rest of the region.
The report asks higher-capacity countries that use datasets from lower-income partners to share the benefits and license the data on fair terms, to avoid extractive practices. One of its headings for these countries reads “Export responsible models, not dependencies.”
Sector playbooks
The report ends with playbooks for eight sectors. They cover health, education, finance, agriculture and fisheries, urban areas and transport, biodiversity and climate, public services and governance, and security and justice. Each playbook sets out uses, minimum standards, guardrails and measures of success.
Health
The health playbook asks that diagnostic tools be at least as accurate as qualified health workers across demographic groups. If a tool reads an X-ray as normal but the patient still has symptoms, a doctor should review the case. When the tool's advice leads to an error, the institution that deployed it should be accountable, not the tool or the patient.
Education
The education playbook measures success by learning gains for the lowest-achieving quarter of students. If AI is doing its job, the report argues, weaker students should catch up the most. It calls an AI careers adviser that shows girls nursing jobs but not STEM jobs “unacceptable.”
Finance
AI credit scoring that uses alternative data, such as phone bills, can open loans to people with no formal credit history. The playbook says regulators can cap interest rates on AI-driven microloans. A small business turned down by an algorithm should be able to ask for a manual review and see the main reasons for the decision.
Common threads across sectors
The report draws four common threads from the playbooks:
Involve users in design, and keep a human in the loop for judgment calls.
Protect data and rights.
Measure outcomes, not outputs.
Measure continuously and iterate responsibly.
Who's behind this
What comes next
The report calls AI the region’s next essential infrastructure, like power, roads and schools. Its measure of success is “not smarter machines but more empowered people,” people who can learn, earn, stay healthy and safe, and trust public institutions.
UNDP’s Project Vision plans an AI for Human Development Dashboard and “a regional database analysing the future orientation of national strategies” to track how countries are doing. The gains the report describes, from rural clinics catching disease earlier to students learning in their own language, already exist in pockets. Whether they reach everyone depends on choices made now.
Pedro Conceição
Director of the Human Development Report Office, UNDP
Published by Center for Human Progress (US) Inc., a 501(c)(3) organization