Education · Idee
Using is not understanding
AI literacy is currently sold as the ability to operate: prompts, tools, courses. The part that counts begins after that — with the judgment about the answer.
In four sentences
- Prompting is a matter of operation, learned in an hour. Judgment is the real competence.
- AI answers are convincingly phrased whether or not they are right — which makes domain knowledge more important, not less.
- If you don't know a field, you cannot check the output — and you adopt mistakes with full confidence.
- Good AI training therefore practices checking, not phrasing.
The wrong lesson
Most of what is being sold as AI training right now is operation: phrasing tricks, tool overviews, template collections. That is not wrong; it is just finished quickly. The models are getting better month by month at making sense of badly posed questions anyway.
What does not get better on its own is the ability to judge an answer. A model delivers its best attempt — linguistically confident, well structured, in the same even tone whether it is right or wide of the mark. The packaging carries no information about the quality of the content.
„If all you can do is operate the tool, you won't notice when you're being lied to — politely, fluently, and fully formatted."
Why domain knowledge is becoming more important
From this follows something that turns many debates on their head: AI does not make domain knowledge obsolete; it makes it more valuable. If you know a field, you can tell in seconds whether an output is usable, and you save enormous amounts of time. If you don't, you adopt mistakes with full confidence — and only notice once someone else finds them.
For teams this means: AI lifts strong people far more than weak ones. The gap grows; it does not shrink. Ignore this and you will later be puzzled by quality problems that cannot be attributed to any one person.
What I would practice instead
A training session that accomplishes something looks roughly like this: real outputs from your own day-to-day work, some with built-in errors, and the task of checking them and giving reasons for your verdict. Then the question of which tasks you will hand the system in the future and which you won't — and what you base that on.
That is more uncomfortable than a prompt collection and considerably more useful. In the end, the point is not working with AI. The point is to still know what you are doing.
Questions about this
What does AI literacy mean in concrete terms?
The ability to judge results: recognizing whether an output is right, where it is uncertain, and which tasks a system should be given at all. Operation is the smaller part.
Does AI make domain knowledge obsolete?
No — it makes it more valuable. Only someone who knows a field can recognize a convincingly phrased but wrong answer for what it is.
What should corporate AI training look like?
Built on real outputs from people's own daily work, including flawed ones, which are checked and assessed with reasons given — practicing checking rather than phrasing.
