Noisy ← this week

The world is being quietly rearranged by people who write very long documents.

March 16, 2026
World Bank
The title they went with
Disruption without Dividend ? How the Digital Divide and Task Differences Split GenAI’s Global Impact Noisy translates that to

Poor countries get AI's layoffs before AI's benefits — the internet gap explains why

The workers most vulnerable to AI displacement in developing countries have enough connectivity to be automated out of a job. The workers who could use AI to become more productive do not have enough connectivity to open the tool. The internet arrived just far enough to remove jobs and not quite far enough to replace them.
What happened
The World Bank ran the numbers and found that AI will displace workers in poor countries before it makes them more productive. Workers vulnerable to automation have internet access even in low-income countries, so they'll feel job losses immediately. Workers who could benefit from AI tools face broken digital infrastructure that blocks them from using those tools at all.
before developing countries assumed to face lower AI risk
after disruption arrives first, productivity gains may not
Why this matters
For years, development agencies assumed that once AI tools existed, poorer countries would eventually catch up to rich ones. This paper shows the opposite happens first: the disruption arrives on schedule, the productivity gains do not. Workers in poor countries will experience automation without the infrastructure to benefit from augmentation — a one-way door. The asymmetry is structural, not temporary. It means development banks and AI-for-good programs built around 'access equals benefit' now have a real problem: access without infrastructure is just job loss.
Who wins, who loses
who wins Workers in rich countries with fast internet — AI tools make them faster at the analytical work they already do.
who loses Workers in poor countries who have just enough internet to be automated out of a job, but not enough to benefit from AI productivity tools.
The signal
Why this hasn't landed yet
The finding contradicts the broadly held and institutionally convenient belief that AI will benefit developing economies by lowering barriers to skilled work. That story supports existing technology investment narratives and is easier to fund. A finding that says the disruption arrives first and the benefits may not is harder to attach to a project proposal. The paper is also a working paper with a disclaimer on page one saying it does not represent World Bank views, which gives editors a structural reason to treat it as preliminary.
What happens next
Development agencies and multilateral lenders that built AI optimism into their workforce investment models now have a documented case that those models used the wrong occupational exposure data. Expect pressure to revise country-level AI readiness assessments, particularly those tied to digital infrastructure lending. The paper's authors are reachable at a World Bank email address, which suggests this finding is meant to travel inside the institution before it travels outside it. The more immediate consequence is for governments in low- and middle-income countries that have been told AI will be an equalizer: they now have a World Bank citation saying it may not be, and may be worth using in negotiations over digital infrastructure financing. That argument gets more useful as AI displacement becomes more visible in tradeable-service sectors, probably within the next two to three years.
The catch
Working papers carry an explicit disclaimer that they do not represent World Bank policy. That disclaimer exists for a reason. The finding that standard exposure measures are wrong is methodologically interesting but operationally inconvenient: it requires rebuilding the measurement tools that development agencies have already incorporated into investment frameworks. Institutions rarely move fast to invalidate their own prior analysis. The paper also hedges carefully, noting that lower task content in developing countries may itself reflect underinvestment rather than structural difference, which gives skeptics room to argue the findings are temporary. No context research was available; these observations are inferred from the document alone.
The longer story
The longer arc
This sits inside a fifty-year argument about whether technology transfers close or widen the gap between rich and poor countries. The internet was supposed to be the great equalizer in the 1990s. Mobile banking was supposed to be the equalizer in the 2000s. Each wave produced real benefits and also produced new infrastructure requirements that the previous wave had not created. GenAI requires bandwidth, compute, and non-routine analytical task content. The World Bank is now documenting that the same asymmetry has appeared again, one wave earlier than most analysts expected.
Part of a pattern
This is part of a growing body of research finding that AI benefits are not uniformly distributed even within countries, let alone across them. The pattern across recent work is consistent: AI augments workers who already have the skills and tools to use it, and displaces workers who do not. The World Bank contribution here is adding the infrastructure layer explicitly and showing it operates differently for displacement versus augmentation, which is a more precise statement of the problem than most prior work.
If you insist
Read the original →
The Sendoff
Workers in automatable jobs in low-income countries have enough internet to be replaced. Workers who could benefit from AI tools often do not. The digital divide turns out to be quite precise about what it allows through.