TL;DR
Instructors at Multiverse, the £1.6bn apprenticeship company co-founded by Euan Blair, told the Guardian that an AI system now marks their online lessons and alerts managers to suspected weaknesses. Several described losing sleep, and some are seeking therapy. Much of Multiverse’s training for the NHS, councils and universities is paid for through the apprenticeship levy, which gives this dispute over workplace AI monitoring a public-money dimension.
How the system works
Transcripts of online classes are fed to an AI model that gives each instructor a “risk status”, a “percentage confidence score” and written comments for managers to scan. It flags a long list of possible faults: taking longer than about a minute to sort out a connection problem, overusing fillers such as “sort of” and “kind of”, giving vague answers, letting learners fall silent, or getting drawn into a long exchange with a single student.
The model’s instructions, seen by the Guardian, say one or two hedges should not count against an instructor; what matters is a “PERVASIVE REPEATED pattern”. A teacher who self-corrects while setting a task should still be marked down, “even if it was eventually straightened out”. Where a manager once observed each instructor monthly, the AI now reviews several hours of their teaching every day.
Two accounts of the same tool
“Since all of this has come in, the stress level has been absolutely horrendous,” one instructor said. Others said managers had queried “digressions” that were really answers to learners’ questions, and one said they had “no insight” into the AI’s analysis, which only managers can currently see.
Multiverse says the tool “directs human time to where it’s needed most” and that only managers write performance reviews. “The most consequential performance management action it can take is to recommend that a human personally reviews a session,” a spokesperson said. The company says staff were told in advance and will get access to the data.
The company, which has more than 800 staff and made a £28m loss last year, was found by Ofsted in February to be below expected standards in several areas, though inspectors praised its instructors.
Looking forward
John Chadfield of the Communication Workers Union wants new laws on how AI is used at work. Whatever happens there, the instructors’ accounts show how such tools can misfire: in one anonymous telling, delivery staff were left out while the system was built, cannot yet see their own scores, and now teach to the rubric rather than to the learner. Multiverse disputes parts of that picture, but for UK employers weighing AI quality checks, giving staff sight of their own data from day one looks like the minimum.