GenAI may not trigger mass job losses in low-income countries, but it risks deepening capability inequality as data, compute and rule‑making remain concentrated in the Global North; investing in data sovereignty, labour‑augmenting applications and Global South governance could steer technology toward decent work.

Main Finding

Generative AI (GenAI) is likely to transform work in the Global South more by reconfiguring tasks and reshaping job quality than by producing immediate mass unemployment. While direct automation risk is lower in low-income countries, structural asymmetries in AI readiness, data/control, and governance create a substantial risk of deepening “capability inequality”: productivity gains and value capture will disproportionately benefit actors with capital, compute, data, and rule‑making power unless decolonial, capability-centred governance is adopted.

Key Points

• Exposure and distribution
  • Roughly 25% of global employment is in occupations with some GenAI exposure; only ~3.3% of jobs are in the highest-exposure category (ILO WP96/140).
  • Exposure is strongly income‑graded: ~34% of employment in high‑income countries vs ~11% in low‑income countries is in exposed occupations.
  • Highest-exposure risk: ~5.5% in high‑income vs ~0.4% in low‑income countries.
  • Women are over‑represented in exposed clerical occupations: 4.7% of female workers (highest-exposure) vs 2.4% of men (global figures).
• Productivity evidence (mainly from high‑income contexts)
  • Field/experimental results show average productivity gains of ~14–15% in customer support (Brynjolfsson et al., 2025).
  • Writing-task experiments find large reductions in time and improvements in quality, with largest relative gains for lower-skilled/novice workers (Noy & Zhang, 2023).
  • GenAI can compress productivity dispersion (benefiting lower-ability workers), acting as a knowledge equalizer for some tasks.
• Structural readiness and concentration
  • Oxford Insights Government AI Readiness Index 2022: global average 44.61/100; North Africa 38.59; sub‑Saharan Africa 29.38. 21 of the 25 lowest-scoring countries are in sub‑Saharan Africa.
  • Major deficits: compute infrastructure, digital skills, R&D, governance capacity—leading to structural dependence on a few firms and jurisdictions.
• Conceptual synthesis
  • Multi‑level model: (1) technological/task layer (substitution vs augmentation), (2) structural layer (AI readiness, infrastructure, data, skills, power), (3) normative/institutional layer (labour law, data governance, participation).
  • Capability outcomes depend on task exposure × structural readiness × governance regimes.
• Risks vs opportunities
  • Short-term: lower immediate risk of mass automation in low‑income countries.
  • Longer-term: high risk of capability erosion through data colonialism, concentrated infrastructure, exclusion from rule‑setting, intensified surveillance/deskilling, and asymmetric value capture.
• Policy/guidance emphasis
  • The authors advocate decolonial, capability‑oriented governance: data sovereignty, investments in labour‑augmenting (not labour‑replacing) applications, social protections, labour rights, and Global South leadership in standards and agenda setting.
• Limitations of the evidence base
  • Heavy reliance on aggregate indicators and studies from high‑income contexts; limited microdata for Global South workers (especially informal sector and non‑English contexts).

Data & Methods

• Research design: integrative, multi‑disciplinary secondary‑evidence synthesis combining quantitative indicators and causal microstudies with decolonial and capability theory.
• Principal quantitative sources:
  • ILO Working Papers 96 (2023) and 140 (2025) — occupational GenAI exposure and employment‑level estimates (including gender breakdowns).
  • Oxford Insights Government AI Readiness Index 2022 — country/regional readiness scores.
  • Experimental/field studies: Noy & Zhang (2023) on ChatGPT for professional writing; Brynjolfsson, Li & Raymond (2025) on GenAI in customer support.
• Qualitative/theoretical sources:
  • Capabilities literature (Sen, Nussbaum), task‑based labour economics (Autor, Acemoglu), and decolonial/data‑colonialism scholarship (Couldry & Mejias; Mohamed et al.; regional Global South analyses).
• Analytical approach:
  • Map theoretical constructs (tasks, capabilities, coloniality) to observable indicators (exposure indices, readiness scores).
  • Descriptive comparative analysis across income groups and regions using published aggregates.
  • Critical interpretive synthesis to derive risks, opportunities, and governance levers.
• Transparency and limits:
  • No new primary data or statistical estimation; numbers drawn from cited sources.
  • Limitations: aggregate data bias toward Global North contexts; under‑representation of informal work, language diversity, and evolving GenAI capabilities.

Implications for AI Economics

• Modeling and measurement
  • Move beyond headline employment counts: economic models should incorporate task‑level substitution/augmentation, changes in job quality (work intensity, monitoring), and distributional channels (who captures productivity gains).
  • Integrate structural readiness variables (compute, data access, skills, governance capacity) and market power indicators into impact models—these condition whether gains are appropriated locally or captured externally.
  • Track capability‑relevant outcomes, not only wages and employment: metrics for autonomy, bargaining power, access to data/value capture, job stability, and participation in rule‑making.
  • Prioritise microdata collection in Global South contexts (including informal sector, language and cultural contexts) and longitudinal studies to capture dynamic effects.
• Distributional and gender dimensions
  • Models must explicitly incorporate gendered occupational exposure (e.g., clerical work) and the risk that GenAI‑driven task changes disproportionately affect women’s job quality and bargaining positions.
• Policy and governance as economic levers
  • Public policy matters for economic outcomes: data‑sovereignty regimes, public investment in labour‑augmenting AI, regulation of data flows, taxation of platform rents, and enhanced labour protections can reframe who benefits.
  • Industrial policies (R&D, skills, local compute capacity) can alter comparative advantage along AI value chains; without them, Global South countries risk being relegated to low‑value, extractive roles.
• Research priorities for AI economists
  • Causal evaluation of GenAI in Global South settings (public services, agriculture extension, micro‑enterprise support).
  • Quantify value capture along AI value chains (data providers, annotation labour, model owners, cloud providers).
  • Study the interplay of AI deployment with labour market institutions (unions, collective bargaining, social protection) and with non‑market capabilities (education, civic participation).
• Normative framing
  • Incorporating capabilities and decolonial perspectives changes welfare assessment: economists should evaluate whether AI increases people’s substantive freedoms (health, education, voice, control over environment) rather than only productivity or GDP growth.

Suggested next steps (for researchers & policymakers) - Expand microdata collection and randomized/quasi‑experimental GenAI deployments in diverse Global South settings. - Develop indicators for capability outcomes and data/value‑capture flows to complement exposure/readiness indices. - Design policy experiments testing labour‑augmenting subsidies, data‑sovereignty mechanisms, and institutional reforms (labour law, social protection) to observe distributional impacts.

Summary takeaway: GenAI can raise productivity and reduce skill gaps for some tasks, but in the Global South the economic outcomes will be heavily shaped by structural readiness and governance. AI economics should therefore broaden its focus to include task‑level dynamics, governance, data control, and capabilities to assess who gains and who loses.