AI and the Digital Divide: Accelerating Inequality or a Lever for Inclusion?
- Stéphane Guy
- 4 hours ago
- 8 min read
Artificial intelligence promises to revolutionize education, healthcare, agriculture, and public services. But that promise carries a condition that rarely gets stated as plainly as it should: you need access first. According to the UNDP's 2025 Human Development Report, only about one in five survey respondents worldwide currently reports using AI, while the ITU's Facts and Figures 2025 puts the number of people still entirely offline at 2.2 billion. AI did not create the digital divide from scratch; it inherited it, amplified it, and gave it an entirely new dimension. The question is no longer whether inequality exists: it does, documented and measurable. The question is what it actually produces—and whether the current dynamics can reverse it.
In short :
A three-tier divide: AI-related digital exclusion operates simultaneously at the individual, territorial, and geopolitical level, with distinct mechanisms at each scale.
The numbers are stark: According to Pew Research Center, 58% of U.S. adults under 30 have used ChatGPT, versus just 10% of those 65 and older. Globally, high-income countries concentrate the vast majority of AI's benefits.
Access is only part of the problem: The real divide is now cognitive and skill-based—knowing how to use AI is not a given, and using it well is rarer still.
Counterintuitive signals exist: In low- and medium-HDI countries, 70% of respondents expect AI to increase their productivity (UNDP, 2025).
Public policy is lagging: Between institutional goodwill and effective on-the-ground deployment, a gaping gap remains.
An Old Divide, a New Layer
The digital divide is not new. It shadowed the rise of the internet in the 1990s, then the proliferation of smartphones in the 2010s. What AI changes is the speed at which gaps widen, and what is actually at stake. Not having internet access once meant being less informed. Not mastering AI tools today could mean being less employable, receiving lower-quality healthcare, and being poorly served by public administrations that are rapidly adopting these systems.
Like every major technological revolution, the rise of AI risks producing deeply fractured societies and exacerbating imbalances between nations on the international stage.
That warning, articulated before the ChatGPT explosion of late 2022, has only been reinforced since. Generative AI accelerated a dynamic already underway, adding a new layer to what researchers call the "cognitive divide", the inequality tied to the capacity to understand, interpret, and navigate increasingly complex digital environments. If you want a grounding in the fundamental mechanisms at play, our AI glossary covers the key concepts driving this shift.
The Three Faces of the Digital Divide in the AI Era
The Generational Divide: A Gap That Looks Different Than You Think
The data is striking: 41% of adults ages 30 to 49 have used it, 25% of those 50 to 64 say the same and 10% of those 65 and older report ever using ChatGPT.*
*34% of U.S. adults have used ChatGPT, about double the share in 2023
But reducing this gap to a simple clash between "digital natives" and seniors would be a diagnostic error. Recent studies consistently show that young people themselves are far from uniformly proficient: many use AI tools without understanding the underlying mechanisms, embedded biases, or real limitations. Data seems to confirm that adoption is highest among younger, better-educated groups, but it is far from universal even within those demographics. We can also see a big difference in AI use between countryes among EU : "Among EU countries, the highest shares of people aged 16-24 using generative AI tools were recorded in Greece (83.5%), Estonia (82.8%) and Czechia (78.5%). The lowest shares were in Romania (44.1%), Italy (47.2%) and Poland (49.3%)."*
*Eurostat, 64% of 16-24-year-olds used AI in 2025
What gets labeled "using AI" can cover vastly different realities: from asking ChatGPT to fix an email, to engineering a prompt system that automates an entire business workflow. Between those two endpoints lies an entire world, and that gap is measured not in age but in education level, professional exposure, and time available to invest in self-training. This distinction matters directly when assessing what AI will actually do to jobs.
The Territorial Divide: AI Does Not Arrive Everywhere at the Same Time
Globally, the gap between connected and unconnected populations remains vast. According to the ITU's Facts and Figures 2025, "74 per cent of the world’s population are online, compared with 71 per cent a year earlier." *
But that leaves 26 % still offline, a vast part in low- and middle-income countries. Even within high-income nations, connectivity is profoundly uneven: globally, 83% of urban dwellers are online versus 48% in rural areas, and "of the estimated 2.6 billion people offline in 2024, 1.8 billion people live in rural areas."*
*ITU, Global Internet use continues to rise but disparities remain, especially in low-income regions
This territorial imbalance reflects a simple economic logic: companies deploying AI tools do so where returns are immediate: in major metropolitan centers and high-value sectors.
Rural areas, post-industrial zones, and underserved communities receive the technology last, if at all. Smaller municipalities frequently lack dedicated digital staff, let alone the capacity to implement complex AI-driven projects. The pattern holds across national contexts, from rural Appalachia and the American Midwest to remote districts across sub-Saharan Africa, South Asia, and Eastern Europe.
The Geopolitical Divide: AI as a New Vector of Domination
This is perhaps the least visible dimension in Western public debate, yet potentially the most consequential in the long run. Low- and middle-income countries risk missing out on the potential benefits of AI due to long-standing digital divide challenges.
The data point that captures the problem most precisely (also flagged by the ILO) is this: in the ILO-World Bank report "Buffer or Bottleneck?", researchers found that across Latin America, "a total of 26-38% of jobs in Latin America and the Caribbean could be exposed to Generative AI." But at the same time, "Up to half of the jobs that could improve productivity with Generative AI – about 17 million jobs – are hindered by gaps in digital access and infrastructure."*
*ILO, Buffer or Bottleneck? Employment Exposure to Generative AI and the Digital Divide in Latin America
The structural dynamics driving this are clear: computing infrastructure, training data, capital investment, and AI talent concentrate in a small number of countries and corporations. China and the United States absorb the lion's share of global AI investment. Europe is fighting to hold a third position. For much of the rest of the world, the question is not one of competition but of basic access to foundational tools.
Jobs at the Heart of the Divide
The most immediate social question remains employment. Here too, recent data resists both catastrophism and naive optimism. Another ILO report (joint study with NASK) finds that 25% of global jobs are potentially exposed to automation by generative AI, rising to 34% in high-income countries.*
The nuance lies in that word: "exposed." Exposed does not mean eliminated. It can mean augmented, transformed, or redirected. But the transformation will not be equally accessible to all. For someone with continuous access to training and upskilling, AI becomes a productivity multiplier. For a low-skilled worker in a sector without structured support, it can accelerate precarity.
This is precisely what we examined in our analysis of which jobs AI will kill, transform, or create: disruption is not uniform, and the most vulnerable profiles are not always the ones you would expect. For a broader framing of these tensions, our deep dive on AI and the future of work maps out the gap between hype and structural reality.
Can AI Also Reduce Inequality? The Other Side of the topic
It would be intellectually dishonest not to acknowledge what the data also suggests: a genuine window of opportunity for populations previously excluded from mainstream systems.
A farmer in Mali or India accessing, via a basic smartphone, an AI tool capable of diagnosing a crop disease* or surfacing real-time market prices gains a productivity advantage that would have been unthinkable without this technology. Adaptive learning platforms, AI-assisted diagnostics in regions without physicians, early-warning systems for natural disasters: these are not marketing promises. They are real deployments.
The Invisible Divide: Skills, Biases, Trust
Access and usage are not enough. A third dimension remains, less covered in the media but equally structural: the capacity to exercise critical judgment over what AI actually produces. The invisible dangers of generative AI (hallucinations, biases embedded in training data, large-scale AI-generated disinformation) are not equitably distributed across populations.
A population trained in digital literacy, accustomed to cross-checking sources, and capable of identifying when a language model is producing unreliable output will be far less exposed to those risks than a population that has never had the chance to develop those reflexes. This holds at both the individual and national level: countries that have not invested in foundational digital education will also be the least equipped to distinguish a genuinely useful AI from a manipulative one. That gap connects directly to AI and automation's real societal risks.
What Public Policies Do (and Don't Do)
The institutional discourse is reassuring in form: the European Union has passed the EU AI Act, the ILO has proposed three policy pillars (infrastructure, technology transfer, and skills investment), and international bodies have repeatedly signaled urgency on digital inclusion.
At the 2024 AI for Good Global Summit in Geneva, ITU Secretary-General Doreen Bogdan-Martin underscored both AI's transformative potential and the imperative of inclusive governance: "In 2024 — in the age of AI and unimaginable opportunities — one-third of humanity remains offline, excluded from the AI revolution, and without a voice,“ Ms. Bogdan-Martin stated. “This digital and technological divide is no longer acceptable.”*
But between stated intention and operational deployment, timelines stretch and resources fall short. The real lever remains education—and that is a decade-long investment, not a quarterly fix.
What This Means in Practice
The digital divide in the AI era is not an access problem solvable by broadband rollout and online courses. It is a power problem: the power to decide how tools are designed, for whom, with whose data, and according to whose values. As long as the teams building these systems remain unrepresentative of the populations they are supposed to serve, AI will remain for most of the world a technology that acts on people rather than one they genuinely use.
This is a political question in the strictest sense. And as such, it deserves to be stated plainly: the democratization of AI cannot be decreed. It must be built, patiently, through concrete choices in training, infrastructure, governance, and representation. AI can reduce the digital divide or deepen it. It will do whatever we collectively decide it to do—provided, still, that the decision genuinely belongs to us.
FAQ: Frequently Asked Questions
What is the AI-related digital divide?
The digital divide in the AI era encompasses all inequalities in access, use, and mastery of artificial intelligence tools. It operates at three levels: between individuals (varying by age, education, and income), between territories (rural vs. urban), and between nations (high-income vs. developing countries).
Does AI actually worsen inequality?
Data through shows a clear trend toward widening inequality between wealthy and poorer nations. At the individual level, the answer is more nuanced: AI can also reduce certain forms of inequality, particularly in access to education and healthcare in developing countries, provided basic infrastructure is in place.
Which jobs are most exposed to AI's digital divide?
According to the ILO, 25% of global jobs are potentially exposed to automation by generative AI. The most vulnerable positions are those centered on repetitive, predictable tasks, held by workers with limited access to continuous training and upskilling.
What is being done globally to bridge the AI divide?
The EU has adopted the AI Act, establishing a risk-based governance framework. The ILO has articulated three policy pillars: infrastructure development, technology transfer, and skills investment. The ITU's AI for Good platform mobilizes multilateral cooperation. However, the gap between declared intent and effective ground-level implementation remains significant.
Can AI help reduce inequality?
Yes, under the right conditions. Adaptive learning platforms, AI-assisted medical diagnostics, and productivity tools accessible via basic smartphones represent genuine opportunities for underserved populations—provided parallel investments are made in infrastructure, training, and data governance.
What is the "cognitive divide" in the context of AI?
The cognitive divide refers to inequalities in digital comprehension and critical reasoning. In the AI era, it manifests as an inability to identify errors, biases, or manipulations produced by automated systems—a risk that is disproportionately higher for populations without solid digital literacy. Understanding how AI systems like neural networks actually work is one concrete step toward closing that gap.