Report alleges bias in ExamSoft facial recognition feature, proposes improvements
The authentication feature, known as ExamID, is used to authenticate the identities of students while they take examinations remotely; but the report authored by Law Professor Gabe Teninbaum notes that the results of the facial recognition matching are often discriminatory especially against students of color.
Professor Teninbaum’s report, which is expected to be published in the Journal of Robotics, Artificial Intelligence and Law, highlights failures to accurately recognize faces of color as a major problem with ExamID.
ExamID uses a form of facial recognition technology to match an image of a student in its database to an image the student submits immediately before an exam. In theory, ExamID then biometrically matches the stored image with the second image as a means to confirm that the correct person is taking the exam, a portion of the report’s abstract explains.
It however regrets that “unfortunately, just as has been the case with other efforts to use facial recognition technology, the use of ExamID has been error-prone and linked to incidents of students of color being unable to access exams in a timely way.”
Professor Teninbaum of the Suffolk University Law School decried the negative effects of ExamID’s face-matching errors in an interview with The Verge.
“These are students who are about to take a high-stakes exam with a lot on the line, and that is very unwelcome. Any time you go into an exam, you just want to focus on the exam…You don’t want to feel like you have these added challenges. Students deserve to feel that their institution is doing what it can to protect their rights, interests, and dignity,” he said.
As part of a way out of the problem, Professor Teninbaum suggests that schools using ExamSoft could make available an identical generic baseline image for every student, and also that the software should enable what he calls a deferred identification feature so that test takers can have their identities verified, even manually, after taking the exams.
The identification deferral, the report adds, could be designed as a feature “…by which institutions can simply toggle on/off, thereby bypassing ExamID until such time that the technology matures into one that does not discriminate.”
ExamSoft has now enabled deferred identification by default, and told The Verge it is attempting to mitigate the challenges associated with face biometrics by giving exam-takers control over their reference images (or “baseline picture”) and allowing its validation feature to be tested by students prior to the exam.