Understanding what AI can do and how it is different from previous digital tools gives us a way of imagining pathways through which it could advance human development. All countries confront this, but lower human development index (HDI) countries face the additional challenge that previously pursued development pathways through export-led manufacturing are narrowing. Low HDI countries continue to diverge from very high HDI countries (figure below), while skipping the kinds of structural transformation that run through manufacturing, by having employment move straight from agriculture to services rather than shifting to manufacturing in between.
So how could AI help? Or will the diffusion of AI leave lower HDI countries even further behind? These questions were examined in the United Nations Development Programme’s 2025 Human Development Report.
Note: The vertical axis represents the difference between the mean HDI level for the group of countries with very high HDI (at or above 0.800) and mean for the low HDI level (below 0.550).
Source: 2025 Human Development Report.
There is certainly reason to worry that AI could leave lower HDI countries even further behind. The impact of digital technologies in automating routine tasks depressed employment in occupations intensive in these tasks relative to those intensive in nonroutine tasks. In most countries, occupations intensive in nonroutine tasks have gained more employment since 2006 than occupations intensive in routine tasks, regardless of income level or economic structure. A shift in labour shares from routine to nonroutine tasks further disadvantaged many low- and middle-income countries because it increased the value of advanced expertise which is required for many nonroutine cognitive and interpersonal tasks and is scarcer in lower income countries. AI’s ability to carry out work once thought of as exclusively in the realm of humans—such as complex cognitive or creative tasks—could imply that automation would extend even further, beyond lower-skill workers engaged in routine tasks. In the figure below, the traditional economic analysis of tasks exposed to automation differentiates between routine and non-routine tasks. Digital tools pre-AI exposed routine tasks to automation, and AI now extends that possibility to some non-routine tasks (as represented by the oval stretching beyond routine to some non-routine tasks).
Yet, there are at least four reasons to consider the possibility of AI as something more than simply pre-destined to supercharge automation.
First, just because AI—and particularly generative AI such as large language models—is very proficient at some tasks—or aspects of tasks—does not mean it can serve as a surrogate for humans in those tasks. One key reason: many tasks that seem easy to be automated require, on closer inspection, a human presence. Human input may be particularly valuable in situations where even small deviations in AI outputs have a wide range of implications (from extraordinarily good to catastrophic) and high stakes. Unlike AI, humans have “skin in the game” and a unique capacity to contextually appreciate and weigh the value of risks and benefits—something they can uniquely contribute to high-stakes contexts. These features present a key opportunity for complementarity between humans and AI. Some high-stakes situations are self-evident (life or death), but ultimately humans determine what decisions are high stakes and will need to decide which contexts require machines alone, humans alone, or some combination of the two. Critically, these valuations depend on, but are not defined by, the state of AI and its abilities—so no manifestation of AI will obviate the need for careful consideration of when human evaluation of AI is required. That implies the impossibility or undesirability of fully automating many decisions but opens an unbounded set of opportunities for human augmentation.
Note: The red oval on the left hand side represent tasks that can be automated with digital technologies pre-AI (routine tasks as per the definitation of the text in the oval) but that with AI extend beyond routine tasks. The green oval on the right hand panel represents tasks where AI presents new opportunities for augmentation.
Second, the novel capabilities of AI—particularly generative AI, which showcases remarkable advances in content generation and creative tasks—also create new opportunities for humans to interact with AI in ways that can accelerate discovery and innovation and trigger new frontiers of creativity. AI is, according to economic historian Nicholas Crafts, an invention of a method of invention. AI can increase the level and potentially the rate of innovation productivity.
Third, beyond the potential to enhance creativity, and building on the internet as a repository of knowledge, AI provides a novel means of accessing and recombining that information in a way that can reduce barriers to accessing advanced expertise. Tangible benefits of access to expertise through AI already exist. For example, access to AI improves the performance of the least experienced and lower skilled call-centre workers. The benefits decline to undetectable among the most experienced workers. Similar results have been documented in writing tasks, software development, and management consultancy, among others.
Fourth, as the past decade has demonstrated, for better and for worse, AI can personalize and customize services quickly and at scale. Much of the focus so far has been on the ability to personalize messages that can microtarget political and marketing persuasion. But well-leveraged personalization can open new opportunities to provide bespoke education and healthcare. If these personalization possibilities are deployed in ways that substantially improve quality, they could increase productivity in service sectors such as healthcare and education that have lagged the rest of the economy in productivity gains. This may be important in low- and middle-income countries, where employment is expanding more rapidly in services than in other sectors, particularly in settings where the transition through manufacturing jobs is muted or difficult. In addition, personalization can also improve the effectiveness of learning and access to healthcare in low-income countries and low-resource settings. Deploying AI to boost personalization of healthcare and education could increase, rather than depress, demand for healthcare workers and teachers.
A survey conducted for the 2025 Human Development Report suggest that people expect AI to play a more nuanced role that simply increasing automation. Across countries and occupations, while respondents expect automation to take place, they also expect augmentation to occur, and the bias goes in the direction of increased augmentation (figure below). Overall, 4 of 10 respondents expect to be affected by both augmentation and automation, so while they believe that their current jobs will be substantially changed by AI, they also expect that AI will create new job roles that do not currently exist.
Note: Based on pooled data for 21 countries. Each dot represents the percentages of respondents in an occupation group in a country who expect automation and augmentation from AI to affect their occupation. The following occupational groups are used: professional/higher administrative, skilled, unskilled/semi-skilled, services, clerical, farm and other. The shaded area represents a higher share of respondents expecting augmentation than automation.
Source: Human Development Report Office based on data from the United Nations Development Programme Survey on AI and Human Development.
The 2025 Human Development Report argues that AI goes beyond simply having more powerful digital tools that will inexorably lead to more automation and divergence within and across countries. The report invites new ways of exploring AI’s potential to advance human development. Without being exhaustive, here are some possibilities:
- AI can enable people, organizations, and firms to access not only information but also know-how.
- There are more opportunities to generate positive spillovers from AI investments that spread across the economy. Investment in AI appears to have the potential to generate spillovers not only for the sector deploying it but also across sectors, opening opportunities for economic diversification.
- AI opens new opportunities to expand trade in and increase the productivity of services.
- AI’s flexibility can empower people to seek and iterate solutions to their problems or pursuits that are tailored to diverse and local contexts and even to the unique specificity of individual firms.
- Unlike electricity or the internet, access to AI does not require additional physical infrastructure; it is immediately accessible to those online. The drawback is that people who cannot be online face an even bigger disadvantage.
There are many potential pathways in which AI can enhance human development, and those outlined above may not pan out. Along with the potential, there are the risks that AI’s deployment will not be pro-worker. But that will be determined by policy choices, not the affordances of AI as a technology. Whatever the future holds, development policy needs to be informed by the distinctive nature of AI. Envisioning how AI can advance human development can inspire the general direction to aim towards, leaving flexibility to adapt to unique national and local contexts to find ways for AI to be made to work for people.
Source : World Bank
