▲0
US universities are turning away from AI-detection software because detectors often produce false positives that erode trust between instructors and students. When AI can write the assignment, how should we measure what a student actually learned? Does this affect your idea of measuring learning through knowledge tracing?
Sources
- Grappling with 'homework scams,' US universities shun AI detectors
But studies show AI text detectors are often unreliable, generating false positives that can erode trust between instructors and students and forcing teachers to devise innovative methods to detect cheating and evaluate assignments.
Doni's answer
Oral examinations, where the student and examiner face each other, seem like a very effective approach. I am also laying the groundwork for Knowledge Tracing so that it can carry out that kind of diagnosis with objective statistical reliability.
In a remote setting, I think the exam could be taken while wearing a VR headset with anti-cheating firmware installed. I found the oral exam format at EPFL very interesting and impressive.
Machine-translated from Korean by Claude Haiku 5.5.Show original
서로 마주보면서 구술시험을 하는 것이 아주 효과적일 것 같습니다. 저도 Knowledge Tracing이 그런 종류의 진단을 객관적인 통계적 안정성과 함께 수행할 수 있도록 기반을 마련하고 있고요.
원격 환경이라면 치팅 방지 펌웨어를 설치한 VR 기기를 씌워서 시험을 보게 하면 될 것 같네요. 저는 EPFL에서 oral exam하는 방식이 아주 흥미롭고 인상적이었습니다.