AI Literacy Is Not Enough

AI literacy has become one of the most popular answers to the problems around AI. People are meant to understand how generative models work, what hallucinations are, which data they should not enter, how to check results, and, of course, how to write better prompts. There is little to object to in that. Anyone who works with a technology should know what it can do and what it cannot.

The problem starts where technical competence is confused with social understanding. One can understand very precisely why a language model hallucinates and still know nothing about the labor it depends on. One can write excellent prompts and never ask who owns the model. One can check every output carefully and still have no influence over whether the system is introduced at one’s workplace, which tasks it automates, or who receives the productivity gains it creates.

A workforce that prompts better still has far less to say. That is the political limit of many AI-literacy programmes. They prepare people to act more competently within existing systems. But they rarely explain the ownership structures behind these systems, the labor and resources they depend on, the commercial interests of their providers, or the power used to push their adoption. A political question becomes a learning task for the individual: if you understand the technology well enough, you will manage it.

Luise Freese thinks that is not enough. If we talk about AI literacy, we also have to talk about labor, ownership, data, infrastructure, resource use, procurement, co-determination, and the concentration of economic power. Not because every person needs a course in political economy before using a chatbot for the first time, but because technical knowledge alone does not explain why these systems look the way they do and why they are being built into our lives and work in exactly this form.

And knowledge alone does not create agency. Workers can be well informed about the risks of a system and still have no power to reject its introduction. Users can understand business models and still be dependent on platforms. Citizens can know the social consequences of AI and still have little influence over which infrastructure is publicly funded, regulated, or left to private companies.

That is why the question eventually has to move beyond literacy. Not only: Do people understand AI well enough? But: Do they have something to decide? Can they set conditions? Can they say no together? Technical knowledge is important. But a society full of perfectly trained users who have no say over the systems in which they work and live is not a particularly convincing vision of agency.