Glasgow researchers trial AI tool that records classrooms

TL;DR:

• Glasgow researchers are trialling Deliberative Instructional Agents (DIAs), an AI tool that records, transcribes and attributes classroom conversation to give teachers feedback within about 100 seconds of a lesson ending.
• The developers stress the tool is a “mirror” for teacher self-reflection, not oversight or correction, and is built so it cannot be used to monitor staff.
• Pupils scan a QR code on existing devices rather than using bespoke hardware; because voices are biometric data, teachers can delete audio after transcription, with consent forms signed by parents and staff.

A four-year research project at the University of Glasgow and The Glasgow Academy is testing whether AI can compress the slow feedback loop of teacher development. The DIAs surface four to six “critical moments” per lesson — points where a teacher’s response shaped its direction — almost immediately, rather than weeks later at an in-service day when, as co-developer Dr Thomas Cowhitt puts it, those moments “have faded from memory”.

Designed around its sensitivities

The project leads are unusually direct about the privacy questions recording children raises, and have built the tool accordingly. Pupils need no accounts or personal information; they scan a QR code on their existing one-to-one device, which acts as a microphone, and identify themselves with a first name or agreed identifier. Because voice is biometric data, teachers can delete audio once transcription is complete, though audio is retained during the research phase. Parents and staff at participating schools have signed specific consent forms.

The framing is deliberately defensive against edtech’s poor reputation. Both developers — who speak as researchers, parents and educators — were pointedly critical of technology “developed by software developers without the input of educators”. Dr Matt Gibson, rector of The Glasgow Academy, insists the value “is not in this AI doing clever things for you” but in channelling human reflection, and that the model is structured so it cannot become a surveillance tool for managers.

Looking forward

The DIAs are being trialled in two all-through schools, one in Scotland and one in England, and point to a more measured strand of UK education AI than the headlines about chatbots and cheating. The unresolved questions are familiar: whether “privacy by design” survives a wider rollout, how consent scales if recording becomes routine, and whether parents accept the trade-off. If the developers can demonstrate the safeguards hold, the project offers a model for introducing AI into sensitive settings on the front foot rather than after a backlash.