False positives, bias against non-English-speaking students and legal proceedings have got the better of automatic detection. Institutions are falling back on oral and proctored exams.
At least a dozen American institutions have deactivated or banned IA-generated text detection tools, including Yale, Vanderbilt, Johns Hopkins, Indiana, Northwestern, Georgetown and New York University. The reason is technical as well as legal: high false positive rates, systematic bias against students whose English is not the mother tongue, and legal action by students accused of being wrongly accused.
Assess process rather than rendering
The answer is not to find a better detector, but to change the purpose of the evaluation. The leads chosen: rethinking the homework to note the process and the reasoning rather than the finished product, returning to the supervised examination in presentation and the paper exam book, reintroducing the oral, and designing exercises that a model cannot perform instead of the student.
The construction site is far from consensual. While 73% of teachers say they have personally met with a case of academic integrity involving AI, about half of teachers want to change their practices as little as possible. Academic integrity experts such as Tricia Bertram Gallant (UC San Diego), Kevin Yee (University of Central Florida) and Marc Watkins (University of Mississippi) argue instead for an assumed overhaul of the evaluation modalities.
Read the full article on the original source
Go to the source →