AI literacy, not an AI ban

AI output is most convincing exactly when it's wrong — language models are trained to predict plausible sentences, not to verify them. A student who doesn't know the difference hands in work that is tidy and incorrect, and the teacher grading it rarely has time to check every source.

Meanwhile anyone's face can be pasted onto someone else's body in seconds, and one prompt can make a hateful message sound convincing. The line between a joke and real harm has blurred, and few students are taught where that line sits before they cross it.

Minervas splits AI literacy into five areas: how AI actually works and why its answers can be wrong, the data privacy students hand over to AI services, harmful content like deepfakes and hate speech, academic honesty when using AI for schoolwork, and the wellbeing risk of treating AI as a confidant.

Every quiz question and every forum thread maps to one of these five areas — not a decorative label, but the same column value used to filter and report on the data.

True-or-false questions with an explanation shown after every answer — including when it's correct, because the score isn't the point. What's measured is whether a student's reasoning changes once they've read why.

A forum where questions get worked through together, with an AI whose job is to ask back, not to answer. It withholds its own opinion until the student has reasoned through several rounds — and if a student mentions serious distress or self-harm, it stops asking questions and points them straight to a trusted adult and Indonesia's SEJIWA counselling line.

A leaderboard that counts both, so learning on your own and helping someone else earn the same kind of credit.

Every question and explanation is written in Indonesian and English, stored side by side, and can't be published if either language is empty. Students learning in Indonesian don't get the afterthought translation.