Test moderators use AI-generated writing to judge literacy standards
To cut the cost of key stage 2 assessments, a UK government project proposes to use scripts generated by ChatGPT instead of real children's writing as examples supplied to moderators who check that grades are awarded fairly
The UK Department for Education (DfE) is piloting the use of AI-generated writing in assessments to evaluate pupils' literacy standards as they transition from primary to secondary school. This move could result in a 95% reduction in annual costs associated with the process. Around 2,000 moderators are responsible for examining the quality of pupils' writing at the end of key stage 2 (KS2), a significant phase in a child's educational journey in England and Wales.
Moderators typically compare students' work to samples of writing often sourced from real children's compositions. This procedure incurs a cost of approximately £100,000 each year. Beginning this year and next year, a subset of moderators will review AI-generated content produced by ChatGPT's GPT-5 model instead of traditional samples.
The trial is currently limited in scope, with the DfE utilizing the technology to generate three sets of writing for one standardization exercise; a full exercise in 2026-27 and 2027-28 will still involve scripts created by children and obtained under the previous contract. Around 20 seasoned local authority moderation managers will scrutinize the AI-produced material to ascertain its genuineness before its integration into the assessment process.
The department intends to make a decision in spring 2027 about whether to continue utilizing AI-generated samples or revert to procuring them from an external supplier. However, this proposed system has sparked ethical and philosophical debates. Rebecca Clarkson, a researcher at Anglia Ruskin University, questions the use of AI-generated exemplifications, stating that it introduces concerns about altering societal perspectives on acceptable writing standards.
Written by urgent.news from New Scientist's reporting — not their text. Machine-written — it may contain errors, so check the original before relying on it.
