Malta just became the first country to require AI literacy before giving citizens free access to ChatGPT Plus.
Course first. Tool second.
I have been talking to educators this week about AI. None of them are indifferent.
One is retiring soon and his concern is structural. He sees AI as trained on westernized, colonized language patterns that have historically marginalized indigenous and minority voices. Our conversation started with the lack of educational resources for indigenous students, the kind that include native teaching styles and the knowledge of elders. It moved into Black Wall Street. The systemic erasure of economic and cultural infrastructure that competed with the established hierarchy. His concern is not abstract. He sees AI trained on the same patterns that produced those erasures. Not a new problem. An old one, now automated and harder to see.
One told me she is learning Spanish. I mentioned AI language tools. She said she wants a human being in the room. Someone whose first language it is. That was the end of that conversation.
A third connects AI energy consumption to ethics and stays out on those grounds.
A fourth worries about brain atrophy. Outsourcing thinking at exactly the moment when thinking about AI clearly matters most.
None of these responses are wrong. They are all honest reactions to something that arrived faster than most institutions were prepared for.
What Malta Got Right
The sequence matters. Malta decided that literacy before access is the right order. Most systems have left that sequence entirely to chance, tool first, use second, consequences whenever they arrive.
The instinct is correct. Not because a course guarantees judgment. It does not. But because the alternative, handing over a powerful tool to an underprepared user and hoping the preparation follows, has a documented track record at this point. We ran that experiment with social media, search, smartphones, and online banking. We are still living with the results.
What Malta is attempting is different. It is saying: before you use this, know something about what you are holding.
That is not a radical position. It is the logic behind every apprenticeship model in history.
Where the Apprenticeship Frame Matters
There is a version of "just start" that is genuinely right. For most AI tools, in most contexts, the cost of a mistake is recoverable. You try the tool, you get a bad output, you notice it is wrong because you know enough about your domain to recognize it, you adjust. The doing precedes the alignment. The structure forms around the function. This is real and it applies broadly.
But not universally.
Some domains carry consequences that make pure trial and error genuinely dangerous. You do not learn surgery by getting in. You do not learn to fly by taking off. The apprenticeship model exists because certain mistakes are too costly to make alone, and because some knowledge only transfers through proximity to someone who has already made the mistakes and survived them.
Malta's literacy course is an attempt at compressed apprenticeship at population scale. Whether a course produces genuine judgment or merely the certificate is the question no one has answered yet.
The Question Malta Has Not Answered
The educators I spoke with this week carry exactly the kind of knowledge that should be in the room when AI literacy is designed.
The person worried about colonized language patterns knows something about whose voice gets centered and whose gets erased, in curricula, in training data, in the implicit assumptions that become invisible when they are consistent enough. That knowledge belongs in the design of any AI literacy course, not outside it.
The person learning Spanish from a human being knows something about what is lost when the medium changes, about what a live person in the room carries that a tool cannot replicate, and why that matters for certain kinds of learning. That belongs in the conversation too.
The person worried about brain atrophy is asking the right question about what happens to the capacity for independent thought when we outsource thinking systematically, early, and without a framework for deciding what to outsource and what to keep.
These are not objections to AI literacy. They are the content of it.
Malta's sequence is right. Literacy before access. But who designs the literacy course? Whose concerns shape what AI judgment means? The educators I spoke with this week are largely outside the room where those decisions are being made.
Most of them are not.
What Comes After the Course
The other question Malta has not answered is what happens after the course.
A certificate is not judgment. Judgment is what forms when you use the tool, get it wrong, notice you got it wrong, understand why, and adjust. That cycle requires the kind of ongoing learning that a one-time course cannot produce, and it requires the specific knowledge to recognize when the output is wrong, which is domain expertise the course cannot substitute for.
The literacy course is the beginning of the apprenticeship, not the end of it.
Which brings it back to where it started. The educators in my conversations this week are not wrong to be cautious. They are right to ask who the tool was built for, whose knowledge it reflects, what it costs to think with it habitually, and what it cannot do that a human being in the room can.
Those questions are not obstacles to AI literacy. They are AI literacy.