In 1948, the Universal Declaration of Human Rights established education as a fundamental right. The reasoning was clear and has not dated: a person who cannot read, cannot write, and cannot reason within the information systems of their society is not free in any meaningful sense. They are dependent on those who can. They are subject to decisions they cannot interrogate. They are citizens in name but not in practice.
We are approaching an equivalent moment. The information systems that govern credit decisions, health recommendations, hiring processes, educational pathways, legal determinations, and the allocation of public resources are increasingly AI systems. A person who cannot understand these systems — who cannot ask them basic questions, challenge their outputs, or recognise when they are producing errors — is, in a meaningful sense, in the same position as the illiterate citizen of 1948.
They are dependent. They are subject. They are disenfranchised by infrastructure.
"The inability to understand and interrogate AI systems is no longer merely inconvenient. It is a new kind of disenfranchisement."
The AI systems that now shape life outcomes are not hypothetical. They are operational. Across the United States, algorithmic risk assessment tools inform bail and sentencing decisions for millions of people who have no ability to understand, challenge, or appeal the scores they generate.1 Across Europe and the United Kingdom, automated hiring systems screen the CVs of hundreds of millions of applicants, often filtering on proxies that encode historical inequalities without the knowledge of the candidates being evaluated.2
In healthcare, AI diagnostic tools are making — or substantially influencing — clinical recommendations that affect patient outcomes. In education, algorithmic systems are determining which content students see, which career pathways are suggested to them, and in some cases which institutions they are considered eligible to apply to. In financial services, credit scoring models built on AI determine whether a person can borrow money to start a business, buy a home, or weather a crisis.
Each of these AI systems produces consequential decisions. Each of them can be wrong. Each of them encodes the assumptions, biases, and limitations of the people who designed them. And for the vast majority of people affected by their outputs, the systems are completely opaque — encountered as outcomes, experienced as facts, with no legible mechanism for understanding or challenge.
Before arguing that AI literacy is becoming a human right, I should be precise about what I mean — because the term is used in ways that range from the trivial to the genuinely demanding.
AI literacy is not the ability to build AI systems. It is not the ability to write code, to understand neural network architecture, or to tune hyperparameters. These are specialist skills, valuable and increasingly in demand, but not what I am describing.
AI literacy, in the sense that matters for democratic citizenship, is the ability to: understand what an AI system is doing and what its outputs mean; recognise the conditions under which AI systems produce errors or encode biases; ask meaningful questions about an AI output before acting on it; and exercise the right to explanation, challenge, or human review where consequential decisions are concerned.3
This is not a high bar. It is roughly analogous to what we expect of numeracy — not the ability to solve differential equations, but the ability to understand a percentage, read a graph, and recognise when a number does not make sense. We have accepted for generations that numeracy is a prerequisite for meaningful participation in modern economic and civic life. The argument I am making is that AI literacy is now joining numeracy and basic literacy as a foundational competency — the absence of which produces a meaningful disadvantage in every domain of life.
The framing of AI literacy as a human right is not merely philosophical. It has a distributional dimension that makes it urgent rather than aspirational. The populations least likely to have access to AI literacy education are precisely the populations most likely to have their life outcomes determined by AI systems they cannot see or interrogate.
Persons with disabilities face AI systems built without them — and often struggle to use the tools that would help them understand those systems, because those tools are themselves inaccessible. Workers in low-wage service sectors face AI-driven performance management and hiring systems, and are the least likely to have educational access to the AI literacy that would help them navigate or challenge those systems. People in the Global South face the highest risk of being excluded from AI's economic benefits while remaining exposed to AI-driven decisions in healthcare, financial services, and public administration.4
The IMF has documented this dynamic in stark terms: developing economies face lower AI exposure than advanced economies, but also lower capacity to benefit — creating a structural risk of growing inequality between those who can navigate AI systems and those who cannot.5 This is not merely a gap in wealth. It is a gap in agency.
"The populations least likely to have access to AI literacy education are precisely the populations most likely to have their life outcomes determined by AI systems they cannot see or interrogate."
The UNESCO Recommendation on the Ethics of Artificial Intelligence (2021) represents the most comprehensive international statement on this question to date. It explicitly frames AI literacy as a component of the right to education and the right to information — and calls on member states to develop national AI education strategies that reach all populations, with particular attention to those currently excluded.6
The Recommendation is right. But a recommendation is not a right. The distance between what UNESCO has asked for and what exists in most national education systems is measured in decades and in millions of children who are currently moving through schools that have not begun to address AI literacy in any systematic way.
This needs to change faster than it is currently changing. The AI systems that will govern the adult lives of children who are in primary school today are already being designed. The gap between their rate of deployment and the rate at which AI literacy education is being developed is not narrowing — it is widening.
I am not making only a moral argument. I am also making a practical one. AI systems produce better outcomes when the people they serve can understand them, question them, and provide feedback on their errors. The evidence from my own research — and from a growing body of human-computer interaction scholarship — is consistent: AI that is legible to its users is used better, corrected faster when it errs, and integrated more effectively into the decisions it is meant to support.7
AI literacy is not only a democratic good. It is a condition of AI working as well as it can. The argument for making it universal is therefore not purely redistributive — it is also about the quality of the systems we are building and the outcomes they produce for everyone.
The literate society is not better only for those who can read. It is better for everyone — because the exchange of information, the functioning of institutions, and the collective ability to make decisions at scale all depend on a baseline of shared competency.
The AI-literate society will work the same way.