AI could accelerate economic catch-up by cutting the time countries remain trapped in low-productivity states, but access to technology alone will not guarantee convergence. Governments, development partners and businesses must strengthen skills, infrastructure, finance, management and institutions to turn AI into sustained productivity gains.
Artificial intelligence could become a powerful engine of economic catch-up, but the IMF working paper by Patrick A. Imam and Jonathan R.W. Temple warns that access to AI alone will not close the productivity gap between richer and poorer economies. Its central message is simple: countries can buy technology and expand education relatively quickly, but converting those resources into sustained productivity is much harder. For governments, development partners and businesses, the AI challenge is therefore shifting from technology access to productive use.
The 44-Year Productivity Problem
The historical numbers reveal why policymakers should be cautious about assuming that technology automatically creates convergence. Across 1,097 comparable country-period transitions, the average time required for economies to leave the lowest state was 16.1 years for capital intensity, 31.5 years for measured human capital and 44 years for total factor productivity (TFP).
This gap matters because TFP broadly reflects how efficiently economies combine workers, capital, technology and organizational capabilities. Countries have therefore been much faster at accumulating machinery and education than at turning those investments into lasting efficiency improvements.
Income mobility is even more persistent. Economies in the lowest relative-income category had a 94.3% probability of remaining there over the following five years. The estimated average time to leave that position was 87.7 years. These figures are historical measures rather than forecasts, but they demonstrate how difficult economic catch-up can be.
Human capital appears important for mobility. Using a fixed median comparison, economies below the median faced an estimated 79.1 years to escape the lowest-productivity state, compared with 28 years for those above it. The paper does not claim education alone causes this difference. Instead, human capital is treated as an indicator of a broader capability base involving education quality, management, infrastructure, finance and institutions.
AI Can Close the Gap, or Push the Frontier Further Away
The study identifies three ways AI could influence development. It can increase frontier productivity, make existing skilled workers more productive, or reduce the costs of learning, adapting and implementing technologies.
The first two can increase global output without necessarily helping poorer countries catch up. If AI mainly rewards skilled workers, sophisticated firms, reliable data and strong institutions, countries already closer to the productivity frontier could capture larger gains.
The third pathway offers greater potential for developing economies. AI could reduce the cost of accessing expertise, translating technical knowledge, diagnosing problems and adapting global practices to local conditions.
The paper’s counterfactual scenarios illustrate the difference. Under historical patterns, the average time to escape the lowest productivity state is 44 years. Under broad AI-enabled technology diffusion, it could fall to about 28.6 years. After 100 years, around 34% of economies would remain in the lowest state, compared with approximately 48.2% under the historical baseline.
Under unequal AI diffusion, however, the average exit time increases to 58.8 years, with more than half of economies remaining in the lowest productivity category after a century. These are scenarios, not forecasts, but they highlight how AI could either support convergence or reinforce existing differences.
Governments and Development Partners Must Invest Beyond AI
For governments, the findings suggest that national AI strategies should not be judged simply by internet connectivity, AI adoption rates or the number of digital platforms available. The real test is whether these technologies raise productivity.
That requires complementary investment in education quality, workforce skills, reliable electricity, digital infrastructure, data systems, managerial capabilities, access to finance and effective institutions. Governments should identify specific barriers preventing firms and public institutions from converting technology into better economic outcomes.
Developing economies may also gain more from adapting existing AI than trying to compete directly in expensive frontier-model development. Applications addressing local bottlenecks, such as agricultural information, technical translation, public administration, logistics or business diagnostics, could deliver significant gains where information and expertise are scarce.
Development partners should follow the same logic. Financing hardware or digital infrastructure without strengthening human and institutional capabilities could increase AI access without improving productivity mobility. Development programmes should increasingly combine technology financing with skills development, institutional capacity, business modernization and complementary infrastructure.
Businesses Face a Race to Turn AI Into Productivity
For private-sector stakeholders, AI creates opportunities to obtain expertise, analyze information and improve operations at lower cost. Companies in developing economies could use AI to overcome some traditional knowledge constraints without building every capability internally.
But access will not guarantee returns. Firms will need to invest in employee skills, management practices, reliable data, digital systems and workflow redesign. AI should therefore be treated as an organizational investment rather than simply another software purchase.
There is also a risk that AI widens differences between firms. Larger businesses with stronger finances, better data and skilled workers may integrate AI faster, while smaller enterprises struggle with skills and implementation costs. Policymakers and development institutions may therefore need to help smaller firms build the capabilities required for productive adoption.
The paper ultimately reframes what it calls the “Intelligence Divide.” The emerging divide may not simply separate economies with AI from those without it. It could separate countries and businesses capable of converting increasingly accessible intelligence into productivity from those unable to do so. The policy priority is therefore not AI adoption for its own sake, but building the capabilities that allow technology to generate investment, competitiveness and sustained economic catch-up.
- FIRST PUBLISHED IN:
- Devdiscourse
