Check: AI makes mistakes
AI predicts likely words. It does not know facts. It sounds confident even when it is wrong, and it is weaker in Burmese, on local context and on recent events.
In our survey, 89% of staff worried about wrong information, but only 14% worried they don’t know how to use AI. That gap is why checking is a habit, not a one-off.
The CHECK routine
Section titled “The CHECK routine”| Step | How | |
|---|---|---|
| C | Compare facts | Check every number, name, date and quote against the original source. |
| H | Hunt for errors | Ask the AI: “What might be wrong or missing in your answer?” |
| E | Evaluate the language | Is the tone right for the audience? Formal for government partners such as PESAO or MECC? |
| C | Confirm translations | Translate it back into the original language and compare. A native speaker checks anything that goes outside InEd. |
| K | Keep a human sign-off | You are responsible for anything you send, not the AI. |
Common AI mistakes to watch for
Section titled “Common AI mistakes to watch for”- Changed numbers: 71% becomes 81%, or two days becomes three.
- Calculations that follow a wrong number: one wrong figure produces another.
- Invented sources: “According to UNESCO (2024)…” when no such source was given. Check every citation.
- Confident guesses about InEd: numbers, partners or roles it doesn’t actually know.
- Translations that sound fluent but change the meaning, especially in Burmese.