AI’s Job Impact Remains Modest as Developing Economies Face a Skills Gap
ILO research finds gradual labor-market effects so far, but warns that weak digital infrastructure and training could widen inequalities.
Artificial intelligence is reshaping expectations for work faster than it is reshaping employment itself. A new International Labour Organization analysis argues that the effects remain relatively modest even in advanced economies, giving developing countries an opportunity to prepare rather than assume a wave of immediate automation is inevitable.
The November 4 policy brief, AI and Developing Economies, says productivity gains are possible but depend on access to computing resources, reliable infrastructure, skills and governance. Without those foundations, economies that most need higher productivity may capture a smaller share of the benefits.
That distinction matters for business leaders and governments. The question is not simply whether a worker’s tasks can be automated, but whether employers can implement new tools affordably and reorganize work in ways that raise output.
The adoption gap
An AI application needs more than a software subscription. Companies require electricity, connectivity, usable data, cyber safeguards and staff able to interpret results. Smaller firms can struggle to finance those complements, especially where credit is expensive or power supplies unreliable.
The ILO warns that delay in building these capabilities could allow existing divisions between richer and poorer economies to widen. A technological opportunity may therefore deepen disparities if governments focus on attracting AI investment while neglecting broad digital capacity.
The report does not argue that every worker needs to become a programmer. Familiarity with AI tools, analytical judgment and the ability to verify outputs may be more widely useful than specialized model-development skills.
Evidence is still mixed
A separate ILO review of empirical research from June found real productivity benefits in some tasks, but limited evidence of economy-wide job displacement. Worker-reported time savings have not consistently translated into higher measured output, wages or employment.
This means claims of an immediate, universal employment shock run ahead of the data. It does not mean that all occupations are safe or that adjustment will be painless. Entry-level positions, administrative roles and tasks involving routine content may change more quickly than aggregate employment figures suggest.
The effects will vary across industries. A bank may use AI to process information, while a factory may need expensive equipment and reliable production data before similar productivity gains are possible.
Training must follow the workplace
The ILO’s Changing Landscape of Skills report emphasizes cognitive, social and digital capabilities alongside emerging technical roles. Employees need to understand the limitations of automated recommendations and retain responsibility for decisions affecting customers, colleagues and safety.
Vocational providers and universities face an implementation challenge: curricula must change without chasing every short-lived product trend. Companies, meanwhile, need to provide training during paid work rather than assume that individuals can finance retraining on their own.
Policies protecting workers should address access to skills, job quality and the bargaining power of employees whose roles change. Social protection can make transitions more manageable, but should not substitute for investment in productive firms.
A development decision
AI policy should be evaluated by whether it supports better services and more productive employment, not just by counts of data centers or start-ups. Governments can encourage fair competition, practical training and reliable infrastructure while assessing risks to privacy and discrimination.
The ILO’s message is measured: a slower adjustment creates room to act. Whether developing economies use that room will help determine if AI spreads economic opportunity or concentrates it.
