How can generative AI better serve children’s rights?

If you have been following the race to regulate artificial intelligence (AI), you might have noticed a group missing from debates about safe use: children. Young people, up to 18, make up more than a quarter of the world’s population and are the largest cohort of online users, which means they will be most impacted by AI over their life course. Yet, they have no seat at the table.

It is true that AI’s rapid evolution makes it difficult for policymakers to keep up, but it is necessary for advocates for children to take stock of how well these technological advances benefit all children, especially the poorest and hardest to reach.

A roadmap, in fact, already exists that allows advocacy for children in the age of AI: the Convention on the Rights of the Child (CRC). Now in its 35th year, the CRC is the world’s most widely state-ratified human rights instrument, as well as one of the most comprehensive in existence. It spans all types of rights and applies to diverse contexts: in times of peace as well as in situations of armed conflict. The CRC is overseen by the Committee on the Rights of the Child, independent experts who represent all world regions.

That Committee distilled a set of four General Principles: the right to non-discrimination; the right to respect for the views of the child; the best interests of the child; and the right to life, survival and development. These help us consider the big picture: how a government is performing in creating the systems needed to advance the rights of children in their country. The principles are a valuable tool to analyze whether the rights of children are truly protected, respected and fulfilled, for both ongoing and – critically – new challenges.

So, let’s see how the four principles apply in the context of generative AI. The distinct focus of each principle makes this a useful framework for bringing out the key issues children face with generative AI. The second, on the best interests of the child, is unique in the way it influences the other three principles.

Non-discrimination

Human rights instruments and constitutions around the world protect the right to non-discrimination. The CRC principle of non-discrimination relates both to how we promote equality across the board and, also, how we prevent and redress discriminatory practices. In the case of generative AI, the major issue of discrimination comes down to a lack of equal access to both the internet and to digital skill development. This inequity could be labeled ‘digital poverty’.

Making use of AI requires, among other things, internet access. Many low-income countries have leapfrogged landlines and gone straight to mobile phones, but the lack of basic infrastructure – now repurposed in wealthy countries for fast fiber-optic broadband – means many households’ digital access remains limited to mobile data. Discrepancies exist from country to country, but also within countries and even within communities. Low-income countries will have to build this infrastructure from scratch.

For instance, while cables have reached Africa, it will be a long time before high-speed internet is available across the continent. This is partly due to the absence of industry incentive to invest in rural areas, leading to higher intra-country disparities with the urban child having better opportunities than the rural child.

Second to equal access is equal opportunity for every child once online. For this opportunity to be realized, children must be protected from digital harms – such as abuse and exploitation – and taught how to navigate the digital space so they can make the most of its benefits. 

As it stands, because of the data on which generative AI is trained, children and adults experience similar or worse bias and discrimination using AI tools as they do in their offline lives.

While some businesses have tried to make these tools safe and useful for children, most only strive for the bottom line and will not create these systems in a way that that truly respects and fulfils children’s rights. Nor do businesses sufficiently provide skill-building among children and their parents so they can safely navigate these systems.

Governments around the world, including in low-income countries, have asked for more support to prepare their children, and their wider populations, to reap the benefits of AI. Unless you are lucky enough to be born in a country where the government has taken the lead to incorporate digital skills into the education system, children and families are largely left to fend for themselves. But even in high-income countries, information in languages other than English is not only limited but will often produce different search results.¹

Empowering children in low- and middle-income countries to understand what generative AI can do for them – and hence to make meaningful use of it – is not the responsibility of governments alone. Businesses must also proactively ensure children’s rights are fully respected and supported when using these new tools.²

Respect for the views of the child

Research and experience show that the best way to ensure the responsible design and meaningful use of these tools by children and young people is to seek inputs from children themselves. This will require important investments by both governments and the private sector.

Consulting children allows us to understand their experiences with generative AI and offers them the chance to express their views on an issue that affects them now and in their future. Meaningful engagement with children and young people would help identify what works, what is broken and needs to be fixed, and ensure a tool is child-sensitive.

Children should also be informed – using age-appropriate language – about what data is being collected from them when they use AI tools, together with why and how their data might be used, and what is at stake. They should also have the option to make use of this data themselves.

This kind of meaningful child engagement means considering their young age and evolving capacities, and helping them analyze potential threats, including how AI has boosted the spread of disinformation through synthetic media (deepfakes, for example). Compared to adults, children arguably face a higher risk of their data being manipulated through AI tools, owing to their age and cognitive development. They are unlikely to be as aware as adults about the potential consequences of sharing too much information. These elements are increasingly critical in a world where children’s online experience is completely integrated into their well-being and development. 

Rolling out AI tools, businesses need to be mindful of the risks to children given their relative vulnerability. Age-appropriate applications and settings offer one level of protection, but the private sector must also empower children to better recognize true and trusted versus false information. Governments, through their education systems, are also responsible for raising awareness among children, including being able to differentiate between information generated by a machine versus a human being, and interactions with a machine versus a human being.³

Best interests of the child as a primary consideration

As mentioned earlier, this principle does not exist in isolation. It is a weighing tool because it is a ‘primary consideration’ in decision-making relevant to a child. This tool allows official bodies to determine whether something is in a child’s best interest or not depending on parameters set by states.⁴

When it comes to making this judgment, a couple of persistent myths should be unwrapped. One is that children perceive a boundary between the world online and the world offline, as if they were distinct. This is a misconception. Their experience is fluid between the two spheres. The other myth is that children spending time online is inherently bad. This, again, is not true. Research shows taking a moderate approach – neither too much nor too little – can both better protect them (by allowing time to learn and practice digital literacy) and help them benefit from their connected time. Enjoying the benefits while avoiding the risks of harm that come with digital and generative AI tools is a skill that needs to be taught and practiced.

Learning this skill, however, should not be left to children and parents. Businesses creating these tools and products bear the responsibility of considering the best interests of children when it comes to the design and impact on children. UNICEF recently concluded a four-year study, together with the LEGO Group, looking at the integration of child well-being into the design of online gaming. It demonstrated that you can successfully take this into account in the earliest stages of design and provide a truly empowering experience to children, not only in terms of respecting their rights and providing a good gaming experience, but also by supporting their well-being as part of that experience.

The right to life, survival and development

So far, this article has looked at some of the benefits and risks from children’s digital experiences, mainly with generative AI, in the areas of play, learning and development. But its impacts extend to many more aspects of children’s lives and rights, including improved healthcare and access to education for the poorest children. 

AI systems can help to diagnose and treat illnesses, process vast amounts of data to facilitate research and treatments, and speed up vaccine development, to name a few life-saving interventions. In this sense AI systems impact children’s lives even when children are not using the tools directly. However, investments in preparing for an AI-empowered future – new data hubs, cooling centers, etc. – increase AI's carbon footprint, including in the global south. It will also strain water resources in a world increasingly short of water. While we hope AI will deliver solutions to climate challenges, its environmental consequences threaten to exacerbate a climate crisis that already impacts on children the most.

Linked to this is the US$1.3 trillion ‘investments’ in fossil fuel subsidies spent in 2022. According to the IMF, we “know these subsidies typically give the richest 20 per cent of the population about six times more benefits than the bottom 20 per cent.”⁵ Rather than subsidizing fossil fuels, these resources should be invested in new models to transform the green economy – to be green not only for the climate, but also for human development. The jury is still out as to whether the investments made into AI will also be a game changer for how development is delivered to bring about an end to child poverty.

Education has the power to be a great equalizer by reducing disparities and breaking intergenerational cycles of poverty. Now we hear generative AI holds this power in education, through teacher support, facilitating learning assessments and personalized learning, which could address learning gaps and improve outcomes for children. But it is not about turning to digital tools completely and forgetting about mainstream education. Evidence shows that digital tools are most successful if integrated within the school system. For the student to succeed in school, digital tools must complement rather than replace teachers.

However, while a chatbot or a digital twin for teachers could positively impact a child’s development, they also present tremendous risks. For instance, when dealing with issues around mental health and well-being, human interaction is imperative for children’s development and growth. Assessing the role of AI-augmented solutions is critical: leveraging the opportunities, where appropriate, and maintaining the human interface, where needed.

The potential of AI is clear. Scaling up quality services for children, empowered by generative AI, will help fulfill their right to life, survival and development. Less clear is how governments, the private sector and organizations like UNICEF will be proactive and prepare children with the skills they need for a digitally empowered world.

One of the starkest challenges to the right to life is children’s exposure to extremist groups. This has been greatly facilitated by advances in digital technology and now, enabled further by generative AI. It is important that governments keep track of these dangerous developments. Children may be instrumentalized as actors to advance new and different kinds of warfare, including by using generative AI, to maintain constant disorder as part of ongoing warfare.

The ultimate danger with generative AI is that it one day creates ‘agents’ that only communicate with other AI systems, bypassing humans, in ways that don’t align with human needs and values, or prioritize one child’s rights over another’s resulting in opaque harms and inequalities. This would make it impossible for us to protect all children, and to understand what is truly happening and what comes next.

As some of the most dramatic implications for children and future generations, both of these scenarios must be forestalled.

Conclusion

The Convention and its four principles remain highly relevant for the world today and tomorrow. They can be used to shed light on how major developments and innovations – such as the advance of generative AI – inter-relate with children’s rights. 

These principles call for action, from everyone, and for leadership from both government and the private sector. 

1 See a provocative perspective in ‘Eradicating poverty beyond growth’. Report of the Special Rapporteur on extreme poverty and human rights, Olivier De Schutter.  A/HRC/56/61 of 1 May 2024.
2 Education ADG of UNESCO on Generation AI: Navigating the opportunities and risks of artificial intelligence in education. She also calls for the private sector to step up! https://www.unesco.org/en/articles/generation-ai-navigating-opportunities-and-risks-artificial-intelligence-education?hub=343&mc_cid=fc598a83c8&mc_eid=fcd26ffd3e
3 AI Policy Guidance: "Make AI systems transparent to the extent that children and their caregivers can understand the interaction. Children should be notified in a forthright manner when they interact directly with an AI system, to avoid a situation where they think they are interacting with a human. In addition, AI should not be used as the only input to determine key life decisions that impact children, for example medical diagnoses, welfare decisions or processing school applications, without a human-in-the-loop to make the finaldecision. Children and their caregivers should be notified that AI systems have been used to guide such important decisions."
4 UNICEF is currently working on a tool for business to assess their child rights impact in the digital environment, which will provide much more practical guidance on how this can be done. It is also consulting children and experts to better understand the views, challenges and needs around interpreting the best interests principle in the digital environment.
5 IMF Managing Director’s Keynote Speech at King’s College, Cambridge, 14 March 2024.