AI review fails when tutors cannot see what changed
A three-month study followed eight tutors who checked an AI learning helper. Repeated checks, unclear effects of their changes, and weak ties with learners and peers made review harder. Changes to the screen design had mixed results. This small tutoring study does not show that the same fixes would work in every workplace.
What the source reports
Researchers worked with eight tutors over three months as they checked an AI learning helper. The tutors had to repeat many checks. They could not always tell what changed after they stepped in, and they felt less connected to learners and peers. The team tried changes to the screens tutors used, with mixed results. A review button alone does not make review useful; a tutor needs to see the evidence and the effect of a correction. This small study took place in tutoring. It does not test the same approach in other jobs.
Original source
- Title
- What UI Design Can Do For "Good" Oversight Work: A Three-Month Action Research Study
- Author
- Cedric Faas, Lisa Dargasz, Julius Gabelmann, Felix Jahn, Markus Langer, Kevin Baum, Anna Maria Feit
- Publication
- arXiv
- Date
- Monday, October 12, 2026