Key Details
The IMF working paper Relative Development and the Intelligence Divide: Human Capital, Technology Diffusion, and AI, examines why access to new technology does not always help less-productive economies catch up with global leaders.
Productivity gaps are unusually persistent: Based on historical country data, the model estimates that economies in the least-productive group took an average of 44 years to move out of it.
Capital and education spread more readily: The corresponding estimate was approximately 16 years for economies with the least capital per unit of output and 32 years for those with the lowest measured human capital.
Capabilities affect the pace of improvement: Economies with lower measured human capital took an estimated 79 years to leave the least-productive group, compared with 28 years for those with higher human capital.
AI can produce very different outcomes: In an illustrative scenario where AI benefits spread widely, the estimated period spent in the least-productive group falls to 28.6 years. Where the benefits remain concentrated in better-prepared economies, it rises to 58.8 years.
These are not forecasts: The scenarios illustrate how different patterns of AI adoption and adaptation could affect development. They do not predict when individual countries will catch up.
Technology Can Spread Without Producing Economic Catch-up
The paper’s main finding is that countries have found it easier to acquire machinery, infrastructure and formal education than to achieve sustained improvements in how efficiently these resources are used.
A country can import equipment, expand schooling or obtain AI software relatively quickly. The surrounding capabilities—reliable data, trained workers, effective management, finance, infrastructure and institutions—take longer to develop.
This helps explain why less-productive economies can grow without closing their relative gap with leading economies. If the leaders are also becoming more productive, improvement alone does not constitute catch-up.
Education Matters, but Qualifications Are Not Enough
Differences in the number of skilled and less-skilled workers explain only a limited portion of the productivity gap between countries. Nevertheless, economies with stronger measured human capital were historically more likely to improve their relative productivity position.
The paper interprets human capital as an indicator of a wider capacity to absorb technology. This includes education quality, management, organisational systems, infrastructure and institutions that allow firms to test, adapt and apply new knowledge.
It does not establish that additional schooling by itself causes an economy to catch up, nor does it identify a universal education threshold after which productivity automatically accelerates.
AI Adoption and Productive Use Are Different
The paper distinguishes between three ways AI could affect economies:
Improve leading technologies: AI could push the global productivity benchmark higher.
Make existing skills more valuable: It could disproportionately benefit firms and workers already equipped to use it.
Help knowledge spread: It could make technical information easier to translate, understand, validate and adapt in economies that currently lack expertise.
The first two effects can raise global output without narrowing differences between countries. AI contributes directly to catch-up only when it helps less-productive economies overcome the practical barriers preventing them from using existing knowledge.
Downloading an AI tool is therefore not equivalent to obtaining a productivity gain. Firms may still need to clean their data, train employees, redesign workflows and change how managers make decisions.
Local Adaptation May Matter More Than Model Size
Developing economies do not necessarily need to build the most advanced general-purpose AI models. A smaller, task-specific system may deliver greater value if it addresses an important local constraint—for example, diagnosing equipment failure, translating technical guidance or supporting an administrative process.
Its effectiveness will depend on whether the organisation has reliable data, domain expertise, human verification and the ability to act on the result. These surrounding conditions are part of the technology’s practical value, not secondary additions.
What Is the “Intelligence Divide”?
The IMF paper defines the Intelligence Divide as the gap between access to increasingly affordable AI capabilities and the ability to turn them into sustained productivity improvements.
It is therefore different from a conventional digital divide. Connectivity, computing access or AI adoption may reduce the access gap, while differences in data, skills, management and institutional capacity continue to produce a large conversion gap.
Policy Relevance
The paper does not estimate how AI will affect India. India appears in a sensitivity test because its large population (along with China’s), gives it substantial weight when the authors recalculate global productivity patterns.
Its framework nevertheless suggests that Indian AI policy should assess what happens after access is provided:
IndiaAI Mission: Compute, datasets and AI applications can lower entry barriers. Their economic value will depend on whether firms and public institutions can incorporate them into everyday operations.
MSMEs: Smaller firms may gain access to AI without having suitable data, specialist employees, finance or the managerial capacity to redesign production.
Workforce development: Training should cover not only the operation of AI tools but also domain knowledge, verification, problem-solving and workflow redesign.
Public-sector adoption: An AI-generated recommendation has limited value if officials lack reliable underlying data or administrative authority to act on it.
Measures of success should therefore distinguish between access, initial adoption, regular use, changes in working practices and actual productivity gains.
Follow the Full Paper Here: IMF Working Paper: Relative Development and the Intelligence Divide