Digital literacy stopped meaning knowing how to use a computer some time ago. It came to mean knowing what a link can do, which sources to distrust, and what happens to information as it travels.
AI literacy is following exactly the same path, and most training is stuck at the first stage, teaching people to operate the tools while the thing that matters is what the tools do to the information they hand you.
What AI literacy actually consists of
It is a short list, it is not technical, and none of it requires understanding how a model works internally.
- Knowing that fluency and accuracy are unrelated. A model's confidence carries no information about whether it is right, which is the single most expensive misunderstanding people have.
- Knowing that ambiguity gets resolved silently. Ask something under-specified and you get a confident answer to one interpretation, with no indication a choice was made.
- Knowing what it cannot know. Anything private, recent, or specific to your situation that you did not supply.
- Knowing that it will not tell you it is out of its depth. There is no equivalent of a colleague saying they are not sure.
- Knowing what you are responsible for. Sending on generated work makes it yours, whatever produced it.
Why this is now a general requirement rather than a specialist one
This used to be a concern for people who worked with the tools directly. It is now a concern for anybody who receives work, which is everybody.
A manager reviewing a report, a teacher reading an assignment, an executive being briefed, a customer reading a policy: each is receiving material that may have been generated, and each needs to know what that changes about how much to trust it.
That is the definition of a literacy rather than a skill. It is the baseline required to participate, not an advantage held by specialists.
What good AI literacy training looks like
Most of it is currently a tour of features, which teaches the half that markets itself and leaves out the half that prevents damage.
- Show real failures from your own domain rather than generic ones. The lesson only lands when the wrong answer is about work people recognise.
- Teach the checking habit, not the prompting trick. What did this get right, what would I have to verify, what did it quietly assume.
- Be explicit about where it must not be used, and why. A rule with a reason survives; a rule without one gets ignored the first time it is inconvenient.
- Make it about receiving as well as producing, because most people will receive far more generated work than they create.