The Security Failures of Online Exam Proctoring

Post Syndicated from Bruce Schneier original https://www.schneier.com/blog/archives/2020/11/the-security-failures-of-online-exam-proctoring.html

Proctoring an online exam is hard. It’s hard to be sure that the student isn’t cheating, maybe by having reference materials at hand, or maybe by substituting someone else to take the exam for them. There are a variety of companies that provide online proctoring services, but they’re uniformly mediocre:

The remote proctoring industry offers a range of services, from basic video links that allow another human to observe students as they take exams to algorithmic tools that use artificial intelligence (AI) to detect cheating.

But asking students to install software to monitor them during a test raises a host of fairness issues, experts say.

“There’s a big gulf between what this technology promises, and what it actually does on the ground,” said Audrey Watters, a researcher on the edtech industry who runs the website Hack Education.

“(They) assume everyone looks the same, takes tests the same way, and responds to stressful situations in the same way.”

The article discusses the usual failure modes: facial recognition systems that are more likely to fail on students with darker faces, suspicious-movement-detection systems that fail on students with disabilities, and overly intrusive systems that collect all sorts of data from student computers.

I teach cybersecurity policy at the Harvard Kennedy School. My solution, which seems like the obvious one, is not to give timed closed-book exams in the first place. This doesn’t work for things like the legal bar exam, which can’t modify itself so quickly. But this feels like an arms race where the cheater has a large advantage, and any remote proctoring system will be plagued with false positives.