AI divide: A new fault line we cannot ignore
Blog
The rapid introduction of powerful AI models has triggered a global rush. Governments and private companies are racing to participate in what is framed as an AI revolution. New capabilities promise efficiency, scale and better decisions across every sector. But this moment marks more than another wave of technology.
AI represents a seismic shift in the goalposts of what it means to be digitally included. When adoption accelerates faster than inclusion, existing inequities are not just exposed, they are amplified. In contexts already shaped by acute and chronic digital divides, AI risks deepening gaps that were never resolved for those most vulnerable; children, young people and their caregivers.
Public narratives about AI and digital inclusion already pull in different directions. Optimists point to AI’s potential to save time and resources, particularly for people learning to navigate digital services and for those who support them. Pessimists warn that AI is advancing with limited regard for people already left behind, widening inequalities along familiar lines of income, geography and opportunity. Both perspectives are true. The challenge is that AI is changing the terrain for everyone at once.
Perceptions shape participation
At UNICEF, we recognize that children’s and adults’ experiences of digital technology are shaped by what they understand and believe about its opportunities and risks. These beliefs influence participation in digital spaces long before technology becomes the barrier. Some parents or teachers assume digital skills only matter for children who will study further, not for those expected to take manual jobs. Some families limit girls’ access to technology because they fear online sexual violence. Adults who have experienced scams, harassment, or misinformation may step back from digital services because they feel unsafe. These perceptions shape who engages, who opts out and whose realities are reflected when AI systems are built.
Access still matters, but the picture is more complex than a simple divide between those who are online and those who are not. Yes, people need devices and connectivity to use many AI powered tools. But AI can also be deployed offline, embedded into back-end systems and woven into decision making processes that operate out of sight. Increasingly, AI shapes how people are assessed, prioritized,or denied access to education, health and welfare services regardless of whether they actively use AI themselves. This means the AI divide must be mindful of every child and every community, not just those who can or cannot use online AI tools.
The dangers to social justice are greatest where AI shaped decisions affect people who rely most on public and health services and who are more likely to be digitally excluded due to age, poverty, disability or distrust of digital systems.
Missing data = missing people
This creates a deeper and less visible risk. Digitally excluded people tend to have a lower digital footprint. Their lives, behaviours and needs are often missing from the datasets used to train AI models. When AI informs policy, allocates resources or automates decisions, this absence matters. Models trained on incomplete or skewed data are less reliable and less fair. The dangers to social justice are greatest where AI shaped decisions affect people who rely most on public and health services and who are more likely to be digitally excluded due to age, poverty, disability or distrust of digital systems.
For a child, these dynamics are not abstract. One child opens an AI supported learning platform and finds a patient tutor that explains maths step by step, adapts to their pace, and builds confidence over time. Another child, living in the same country, is met with a warning that the service is unavailable, restricted or requires payment, or sees their questions rejected as unsafe or irrelevant. Over time, these early encounters shape who feels capable, curious and ambitious, and who learns to stay silent or disengage. AI does not just answer questions. It quietly teaches children what is possible for them.
Capability and confidence still matter deeply. Early experiences with AI shape how people see themselves and their place in a digital world. When systems feel exclusionary, biased or misaligned with lived reality, people do not challenge them. They opt out.
Quality gaps widen inequality
Power and cost add another layer. The key issue is not individual tools, but the widening gap between AI models themselves and the agentic architectures built on top of them. Advanced models require enormous computing power, specialized infrastructure, and sustained investment. Those accessed through paid AI tools are already far ahead in quality, reliability and scope compared to most openly available alternatives. As more services rely on agentic architectures to automate tasks, route users or make recommendations, these quality gaps translate directly into unequal outcomes. At UNICEF, we see that these dynamic risks reinforcing divides between institutions and between the global north and the global south.
Equity, safety and lived experience are harder to quantify and easier to ignore.
We lack shared metrics for inclusion and harm
Compounding this is a lack of agreed metrics for inclusion, usability, and harm. Performance is easy to measure. Speed, accuracy, and scale are rewarded. Equity, safety and lived experience are harder to quantify and easier to ignore. Without shared standards, systems can be optimised for technical excellence while quietly excluding those who do not fit dominant data patterns. What is not measured is rarely prioritised.
This is why education on AI, ethics and political economy is essential. People need to understand not only how AI is used, but how it is produced, trained,and governed. The old adage that nothing is free still applies. Many systems offered at no cost rely on data extraction, limited functionality or dependency on external providers. Without understanding these trade offs, communities cannot make informed choices or advocate for fairer alternatives. Digital inclusion now means understanding power, data, and infrastructure, not just interfaces.
There is also a democratic gap, and it matters deeply for children and for the future of democracy itself. The loudest voices in AI debates tend to be those who are digitally engaged, confident, and well resourced. The experiences of digitally excluded families, children and communities are rarely centred, even though decisions shaped by AI will influence education pathways, access to services and how rights are exercised over a lifetime. When these voices are missing, trust erodes. Children grow up in systems they did not help shape, and societies lose the chance to build a shared understanding of what responsible and ethical AI should look like. Over time, this weakens democratic participation and risks normalising decisions made about people, rather than with them.
Why this matters now
AI will not fix poor design or broken systems. It will not automatically simplify complex public services. Yet it is already shaping how decisions are made and how power is exercised. Governments now face a responsibility to govern AI in ways that recognise every child and every community, including those who may never actively use AI but will still live with its consequences.
The AI divide is already forming. Act now and it can be shaped with fairness and care. Wait, and it will harden quietly. The choice is simple. Build AI systems that work for everyone or allow a new generation of inequality to take root unseen.
This is the first of blog series on AI and what UNICEF is doing bridge the digital divide. In the next blog, we’ll explore data sovereignty issues and specialized models on whether they help or hinder governments and their ability to meet children’s needs.
Find out how UNICEF is helping countries bridge the AI divide