EC[ON]OMY

AI Is becoming more accessible. Why aren’t productivity gains?

Access to artificial intelligence is spreading faster than the capacity to use it productively. An IMF working paper examines why the same technology can deliver different economic outcomes – and what it would take for AI to help countries catch up.

Connecting to an AI model takes minutes. Reorganizing a business, training its workforce and making its data usable takes considerably longer.

That gap sits at the heart of the economic debate about artificial intelligence. If powerful tools are becoming widely available, why should the benefits remain concentrated?

Patrick Imam and Jonathan Temple explore this question in Relative Development and the Intelligence Divide: Human Capital, Technology Diffusion, and AI, published in the IMF Working Papers series.

Their central distinction is between access to knowledge and the ability to put it to productive use. Technology can spread across borders without bringing economies closer together.

Resources accumulate faster than efficiency improves

The study examines 1,097 five-year country transitions between relative development states. These are country-period observations, not a count of countries.

Using a common sample and comparable state definitions, the estimated average time to leave the lowest state is:

  • 1 years for capital intensity, measured here through the capital-to-output ratio;
  • 5 years for measured human capital;
  • 44 years for total factor productivity.

Total factor productivity captures how efficiently labor and capital are combined. It helps explain why comparable resources can produce different amounts of output.

These figures describe the persistence of historical differences. They are not forecasts for individual countries. Nor does leaving the lowest state mean reaching the level of the world’s leading economies.

The economic message is straightforward: countries appear able to accumulate capital and measured human capital more readily than they translate those resources into sustained improvements in relative productivity.

Between access and results lies the organization of work

A business can acquire a modern tool while retaining its existing inefficiencies.

Consider a company that uses AI to prepare reports. Documents are produced faster, but inventory records remain fragmented, procurement is delayed and equipment sits idle. Reporting takes less time; the main production bottlenecks remain.

A broader payoff requires changes to the process itself: defining the problem, preparing the data, assigning responsibility and checking the outcome.

Employees need enough subject knowledge to detect errors. Managers must decide which decisions can be accelerated and where human oversight remains necessary. The company needs resources for integration and training.

User numbers and software licenses therefore indicate adoption. They do not, by themselves, establish an economic return.

Two ways AI can affect productivity

The paper distinguishes two mechanisms with different implications for the gap between economies.

The first is augmenting existing skills. AI helps qualified workers and capable firms perform more productively. Engineers analyze technical information faster. Managers prepare decisions more efficiently. Specialists process complex material.

But the largest returns may accrue where professional teams, reliable data and effective management are already in place. Global productivity could rise while international gaps persist.

The second is making knowledge easier to adopt. AI can help translate instructions, organize information, adapt established practices and validate decisions where expertise is scarce.

This mechanism matters particularly for economies seeking to catch up. It could broaden the range of firms able to use productive knowledge. The outcome depends on whether the technology actually reduces the costs and difficulties of learning, adaptation and implementation.

Why professional knowledge still matters

An accessible answer is not necessarily an applicable solution.

Using a technical recommendation requires understanding the conditions under which it works. Checking a calculation requires knowing its assumptions. Adapting an instruction requires familiarity with the equipment and process involved.

Knowing how to operate an AI interface is therefore only part of the required skill set.

The study finds an association between human capital and differences in productivity mobility. It does not establish a universal education threshold beyond which an economy automatically begins to catch up.

Workforce preparation matters. So does an organization’s capacity to make use of that preparation.

Growth does not necessarily mean catching up

An economy can improve its own performance without narrowing its distance from the leaders.

Businesses may increase output and adopt new technologies. Yet if leading economies improve faster, the relative gap can remain unchanged or widen.

AI can accelerate both sides of that process.

Technology adoption, productivity growth and economic convergence should therefore be assessed separately. For a company, the immediate question is whether a particular implementation delivers results. For economic policy, another question matters: how widely successful applications spread across firms and sectors.

What this means for Kazakhstan

A practical implication for Kazakhstan is to start with a specific constraint inside the enterprise.

This is an editorial application of the paper’s argument, not a country-specific assessment by its authors.

If the problem is document processing, measure turnaround times and error rates. In production, examine downtime and output per hour worked. In procurement, track lead times and the frequency of material shortages. In customer service, assess resolution times and repeat inquiries.

The starting position should be recorded before implementation. Subsequent performance should be compared on speed, quality and cost, including integration, training and the checking of model outputs.

How saved time is used also matters. Preparing a document faster delivers a limited benefit if the approval process remains unchanged and decisions are still delayed.

Sharing successful experience should involve more than naming the software. Other firms need to understand the conditions behind the result: which data were required, what employees learned and which management practices changed.

The productivity work begins after connection

AI broadens access to knowledge. The economic return emerges when organizations can verify that knowledge, adapt it and change how work is done.

The central implementation question is simple: what became faster, more accurate or less costly? And did the improvement survive after all associated costs were included?

The answer reveals how far an accessible technology has become a productive economic resource.

Shyngys Yerbolat, EconomyKZ Research expert, exclusively for EconomyKZ.org

Scroll to Top