AI use at SET

1. Students can use AI on their work. We just ask you to be honest about it.
2. SET uses AI only as a student helper — transcribing voice, reading handwriting, light spell/grammar cleanup. We tell you when. Course content is written by qualified teachers, not by AI.
3. We don't run AI-detection. The detectors are unreliable (~60-70% accurate with high false positives). They disproportionately flag non-native English writers, which would be catastrophic for SET specifically.
4. For high-stakes Cert assessments, we use live oral defence — a short video call where you explain your work. That's how we know it's yours.

Anything. Drafting, brainstorming, grammar checking, translation, understanding a confusing concept, getting feedback on your writing, debugging code, you name it.

The only rule:declare what you used and how. Every submission has an AI-use field. Tick it if you used AI, briefly say what you used it for. SET doesn't penalise AI use — we just ask for honesty.

Why we ask: employers, especially in skilled migration pathways, increasingly expect workers to use AI responsibly with attribution. Building this habit at SET prepares you for that workplace reality.

• Submit AI output as your own without declaring it.
• Lie about how much AI you used.
• Pay someone else to use AI to write your work and submit it as your own. (This is impersonation, not AI use, and we treat it accordingly.)

For high-stakes assessments (Certificate II and III final assignments, PEV English-readiness sign-offs), SET requires a short live oral defence. You explain your work to a teacher on video. This is the moment that matters most — make sure you actually understand what you submitted.

• Course content — written by qualified teachers and approved by a second teacher before you see it. AI does not draft lessons, quiz questions or narration.
• Voice-to-text— runs free in your browser (Web Speech API). Your audio is processed by your browser's built-in voice recognition (Google on Chrome, Apple on Safari). SET never receives your audio.
• OCR for handwritten work — runs in your browser (Tesseract.js). Your photo is processed locally on your device. The OCR text is what SET stores.
• Spelling and grammar cleanup— done by an open-source grammar checker (LanguageTool) self-hosted on SET's servers. Text only, no personal identifiers sent to external services.
• Light style polish on transcripts — Claude Haiku, sub-cent per submission. We show you both raw and cleaned versions; you can switch back.
• Teacher tools — Claude assists teachers with translating feedback into gagana Samoa. Teachers write the feedback and always have final say.
• Auto-detection — we run rule-based detection for at-risk students, knowledge gaps, plagiarism between students (5-word shingle matching, not AI), and writing-process signals (paste %, time on task). None of these accuse a student — they surface patterns for teacher review.

• Run AI-detection that produces "this is AI-written" scores. The tech isn't reliable enough.
• Replace human teachers entirely. AI assists teachers, doesn't replace them.
• Train external AI models on your work or personal data.
• Use AI to make consequential decisions about your account (passing/failing, suspending) without human review.

AI makes mistakes. SET's OCR sometimes misreads handwriting. Voice transcription sometimes mangles Samoan-accented English. That's why every AI output passes through your own review — you see what the AI thought, you decide what stays.

We particularly worry about bias against non-native English speakers. If you ever feel an automated SET feature has treated you unfairly, email [email protected] and we'll review.

AI capability changes fast. We'll update this policy as we change our practices, with 14 days' notice for material changes. Previous versions are kept on file.