Skills for an AI world: Where we stand today

“To summarize information or conduct research, honestly, [AI] makes my work much faster, but it does worsen my research skills.” – a 16- to 17-year-old from Mexico.

Rapid developments in Artificial Intelligence (AI) have altered how young people learn, socialize and prepare for work, and brought new and evolving opportunities and risks for children. These changes are only set to continue, for example, through AI’s fundamental future impacts on work.

To avoid such developments widening existing inequalities, we need to understand how best to prepare children to thrive – both today and in the future – in these new environments. This is the goal of UNICEF Innocenti’s new ‘Skills for an AI world’ initiative, which aims to understand the full range of skills children need to be safe, to develop fully, and to be prepared for their working life in a world where AI plays an increasing role. It explores this by consulting experts, young people and, crucially, by engaging children themselves through both research and consultations. The initiative is focused on children in low- and middle-income countries (LMICs), in particular those across the African continent, in keeping with UNICEF’s Africa Strategy.

This article sets the scene for the initiative, sharing four insights on where we stand today in terms of children’s engagement with, and perspectives on, AI. It is informed by a light literature review and two expert consultations conducted by UNICEF Innocenti in October 2025.

1. Children are engaging with AI in increasingly diverse ways

Globally, most young people say they have some existing understanding of AI. It is widely used and increasingly so among younger children: last year, research by UNICEF revealed that 37 per cent of 9- to11-year-olds in Argentina turned to ChatGPT for information. Children engage with it with curiosity, playfulness and enthusiasm for a range of purposes.

In education, AI tutors, chatbots and adaptive systems help to personalize learning. In Brazil, Cetic.br’s study on AI in education found that seven of ten high school students who use the Internet have turned to generative AI tools to help them with schoolwork. During a UNICEF consultation, a 14- to 17-year-old child in Malaysia described how “I like studying with ChatGPT, can do schoolwork and learning [sic] Malay.”

In Mexico, surveyed 16- to 17-year-olds shared that they use AI in education and beyond: while teenagers there use AI primarily to look for information, help with research or homework, they also use it – albeit to a lesser extent – for creative activities. This can include authoring stories or poems, experimenting with new ideas, or generating images and videos. Similarly, in Brazil and Thailand, some 13- to 17-year-olds use it to write, create art and make music, as emerged from child consultations conducted by the Digital Futures for Children (DFC) centre. A Brazilian 16-year-old described this positively: “AI makes me happy because I feel like I can liberate my creativity; it allows me to express ideas that would otherwise be stuck inside of me.”

These uses appear similar to how children engage with the digital environment in general – information and entertainment activities are usually more common, while creative pursuits are less so.

AI is also increasingly used for social and emotional support. On this there is some data from high-income countries. In the United States of America, over half of 13- to 17-year-olds regularly use AI for companion-like interactions, and almost a third of those surveyed find AI conversations as satisfying or more satisfying than human conversations. In Italy, over 40 per cent of teenagers (aged 15–19 years) turn to AI to ask for help when they are sad, feel anxious or for help making important life decisions, according to a recent survey led by Save the Children. Though less is known about the extent that AI is used in this way by children in LMICs, forthcoming UNICEF Innocenti research suggests it is not insignificant. It will be important to understand the short- or long-term impacts of this type of interaction on children’s development.

To succeed today, children need the relevant digital skills to engage with these AI tools, understanding what tools are available, how they work and how best to engage with them. But they also require broader competencies: critical thinking, creative and relational thinking, the ability to evaluate information and learning how to learn. And underpinning this is the basic need to improve learning outcomes.

There are many questions to grapple with here: How much are skills for generative AI similar or distinct to other digital skills or well-established life skills? What are the new skills needed? And how much should we be focusing on foundational learning vs digital literacies? Research efforts into the evolving skills needs of children, and the ongoing updates to digital skills frameworks, reflect the dynamic impacts of the digital transformation and the need to keep interrogating such questions. In future articles, we will delve into these issues in more detail and share some of our emerging evidence and insights.

2. Children’s safety and privacy are overlooked

There are signs that some children engage with AI critically: the same teenagers consulted in Thailand cross-check multiple sources of information to verify outputs. However, more broadly it appears that children over-trust AI tools – a concern increasingly held by parents and teachers. In Mexico, only 8 per cent of surveyed high school students reported regularly verifying the accuracy of information generated by AI tools. At a UNICEF consultation, a teenager in Uganda shared that “Most of us normally trust ChatGPT or it gives some good, correct, okay, relatable answers.”

Beyond the risk of acting on false information, experts in our consultations highlighted the risk to children of safety and privacy violations, surveillance and extractive data practices. Since AI systems are so complex, hyper-personalized and opaque, children may not know what information is safe or unsafe to share, and may under-estimate infringements on their rights. Adding to these safety concerns, the experts also noted increasing mental health challenges that emerge from exposure to inappropriate content and exploitative entertainment systems.

A holistic response is needed, including protective AI systems and education. UNICEF is focused on such an approach through its recently updated AI Guidance on how AI systems can uphold children’s rights, including ensuring their safety and protecting their privacy.

Experts highlighted that children need to understand what happens to their data with AI systems – how it can be used for training models and sold to third parties. This understanding can help them to filter and monitor what they share more intentionally. Children also need to be equipped with the ability to navigate online spaces safely in the same way they are made aware of offline harms and dangers.

3. Children’s voices are absent in AI design and regulation

Though there are positive examples of children’s voices being included in matters related to AI (see the work of the Children’s Parliament in Scotland), UNICEF guidance highlights the broader lack of inclusion of children in AI design. This is part of the absence of children’s voices in governing digital spaces more generally: analysing over 300 digital policies from 35 countries, the London School of Economics and Political Science found that children are rarely consulted on opportunities or risks, and even less so on solutions.

Children want to be included in AI development – in its design, regulation and oversight.

Children have a right to spaces and opportunities to participate in decisions that affect them and to exercise agency. The world of AI is no different. UNICEF’s consultations with children on AI in 2020, and the DFC’s more recent study both found that children want to be included in AI development – in its design, regulation and oversight. As highlighted in Microsoft’s white paper on AI and the future of work in Africa, input from youth is needed alongside that of community leaders, academics and business leaders, to develop inclusive and relevant AI policies. Child participation is also essential as we try to understand how best to prepare children for an AI world.

Informed by experts, and guided by insights from UNICEF Youth Foresight Fellows, our Skills for an AI world initiative aims to centre children’s voices through consultations to understand children’s own perspectives on the skills they need to thrive. In future pieces, we look forward to sharing what we learn from them.

4. AI use occurs in deeply unequal contexts

We cannot fully understand children’s engagement with AI without acknowledging the broader landscape in which this takes place.

Addressing the skills needs of children at the sharp end of inequalities must be recognized as just one component of the wider goal to address the systemic and structural inequalities facing children.

It is true that interest in AI is global, with this being one of the themes of the recent G20 summit hosted in South Africa. Sixteen African countries have already developed national AI strategies and tech giants continue to invest globally. Yet the International Monetary Fund’s AI Preparedness Index shows wealthier economies in a far better position than lower-income countries to support AI engagement, due to higher levels of investment, and digital infrastructure factors such as electricity access, device access and affordable connectivity.

This creates huge regional disparities in terms of AI use: only half of individuals in low-income countries owning a mobile phone and one in five using the internet; these countries account for only 1 per cent of Generative AI traffic. There is similar variation when it comes to skills. At a macro level, youth in LMICs expressed more interest in developing skills related to AI literacy, while those in higher-income countries wanted to develop human skills that AI cannot do, according to Generation Unlimited’s global survey of 330,000 15- to 34-year-olds.

The uneven picture does not disappear when access and infrastructure issues are resolved. AI training data is overwhelmingly based on data from the global North and few, widely-spoken languages: only 25 of the 6,900 languages spoken worldwide are used to train Large Language Models that power many popular generative AI products (we do see efforts to address this, such as through the development of indigenous language AI tools like Masakhane’s Natural Language Processing tool in Africa or TuluAI in India).

However, at this point in time, the result is that Generative AI responses can exacerbate existing inequalities by being largely based on the contexts and experiences of humans living in higher-income countries. Children consulted in Kenya, India, and Thailand expressed their concern about the lack of cultural diversity in generative AI responses. A 16-year-old in Kenya described one situation where “It was supposed to be an African child, but we were given a white child in a green shirt”.

However, AI can also be part of the solution. In India, AI can act as an equalizer by improving access to English-language learning. Thirteen- to 17-year-olds consulted in Brazil highlighted its potential to promote inclusive education practices, for example by creating accessible education resources and tools for children with disabilities. This use case is being piloted in Uruguay through UNICEF’s Accessible Digital Textbook initiative, which employs AI to include narration, image captioning and text simplification in learning materials to make them more accessible to all children.

We must remember that a high level of skills alone does not lead to better outcomes for children, as systems based on biased data will produce outputs that do not represent many children in the global South or of under-resourced backgrounds.

Addressing the skills needs of children at the sharp end of inequalities must be recognized as just one component of the wider goal to address the systemic and structural inequalities facing children, both in terms of their digital access and AI use, but also in their lives more broadly.

What’s next

In the next article, we will consider what the future of AI developments may look like relevant to children. We will also share what we learn from our child consultations on the skills they need for today and tomorrow, and recommendations to ensure these needs are met.

How did we use AI in this article? This was written by the authors, but an AI tool was used to summarize some literature and to help structure the article. All AI-outputs were verified by the authors.