AI & Digital Systems

AI is the newest transition the Lab studies, and the one where its posture, social science first, technical literacy alongside, matters most. The central risk is not technical failure. It is that these tools are deployed into contexts they were never designed for, and either miss the people they should serve or deepen existing divides.

Inside the programme

Data centres and digital infrastructure asks the physical siting question that a policy conversation about AI usually skips. Where the load lands, whose grid firms it, whose water cools it, and whether the anchor-tenant contract that pays for it looks anything like a supply arrangement other industrial users can access. The Lab treats a data centre the same way it treats a smelter: a large industrial customer arriving in a place, with the same distributional consequences to read.

Automation and skills displacement asks what happens on the ground when a benchmarked model is deployed into a distribution the benchmark did not sample. The Lab looks at absorptive capacity, at where the model fails and for whom, and at the ordinary consumer-protection categories that agentic systems quietly collapse. The unit of study is the person on the other side of the decision.

---

The market, in brief

• High-income countries host 86% of the world's top 500 supercomputers (World Bank); Africa holds under 1% of global data-centre capacity for 18% of the world's people.
• Under 5% of people in low-income countries have basic digital skills, against 66% in high-income countries (World Bank Atlas of AI inequalities).
• Middle-income countries are now heavy GenAI users, yet low-income countries are under 1% of global usage (World Bank WDR 2026).

---

What we see that others miss

The World Bank frames AI readiness as four Cs, connectivity, compute, context, competency. The Lab works on the last two, the human and local ones.

A model trained elsewhere may not fit local realities; a service may assume connectivity or literacy that is not there; a tool may shine in a demo and fail in a clinic. The Lab documents whether an AI system is trusted, usable, owned, and acted upon, the questions that decide real-world impact and that pure model evaluation never touches. Our Reef Support case study shows the pattern: the real challenge was human trust, not accuracy.

---

Knowledge we draw on

• World Bank, Digital Progress and Trends Report 2025 and the Atlas of AI inequalities.
• World Bank, World Development Report 2026 on AI.
• IMF, research on AI, compute access, and inequality.
• Our own Reef Support case study.

---

Related reading

For the full series across the transitions we study, see Articles.

---