Test moderators use AI-generated writing to judge literacy standards
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- UK Department for Education is testing GPT-5 to produce writing samples for KS2 literacy standardisation, claiming it cuts the ~£100,000-a-year moderation process by 95 per cent.
- GPT-5 is currently generating three collections of writing for one standardisation exercise, while one full exercise in 2026-27 and 2027-28 will still use real children's scripts.
- Around 20 local authority moderation managers will review the AI output for authenticity before use, and the DfE plans to decide in spring 2027 whether to keep the approach.
- Rebecca Clarkson at Anglia Ruskin University called using non-real children's writing a "philosophical and ethical issue," and said moderators she spoke with objected on the same grounds.
- Jo-Anne Baird at the University of Oxford warned of a "backwash" in which AI-generated samples become the benchmark teachers push pupils to emulate, saying "this could lead us into some strange places."
- The DfE's own risk assessment acknowledged that LLMs "often [exclude] atypical vocabulary and sentence structures that might be used by neurodivergent or non-native [English-speaking] students," with mitigation planned through "thorough review."
- The DfE's transparency record lists no formal impact assessment for the AI system, and the department did not respond to New Scientist's request for comment.
Why it matters: KS2 moderation directly shapes the level of support ~2,000 moderators recommend for pupils entering secondary school, and the DfE's own paperwork concedes AI text tends to omit the language patterns of neurodivergent and non-native English speakers — the very students whose support calibration is most sensitive. No formal impact assessment is on the transparency record.
Ask SkimNews
