All insights

Weekly4 September 20261 min read

This week in AI: back to school, and the trouble with detectors

As schools, colleges and universities return, the debate about AI and assessment continues. The evidence points away from detection tools and towards better-designed tasks.

What happened

The new academic year starts with familiar questions. Over the summer, Ofqual warned that coursework will need more scrutiny to protect the credibility of qualifications. The Higher Education Policy Institute published pieces arguing that many assessments measure outputs AI can now produce, and that AI detection tools flag international and second-language students more often than others. Google is offering eligible students a free year of its AI plan.

Our take

AI detectors are tempting because they promise a simple answer. The evidence suggests they are not reliable enough to base decisions on, and that their mistakes fall unevenly. A tool that wrongly accuses one group of students more than another is not a neutral safeguard.

The better answer is to design the task so that the thing you want to assess is visible: oral explanations, drafts and revisions, work done in the room, or problems that require judgement about a specific situation.

The same lesson applies beyond education. Public bodies that receive applications, bids or consultation responses will see more AI-assisted writing. Trying to detect it is likely to be unreliable and unfair. Asking questions that need specific evidence, and checking that evidence, works whoever or whatever did the writing.

Keep reading

Related insights