Proctorio, an exam surveillance software, used facial detection technology based on OpenCV that failed to recognize Black faces over half the time, causing potential academic consequences for students of color.
Proctorio is exam surveillance software designed to monitor students during online tests to prevent cheating. The software relies on OpenCV, an open-source computer vision program with a documented history of racial bias issues. A student researcher, Akash Satheesan, discovered that Proctorio's facial detection failed to recognize Black faces more than 50% of the time, while achieving less than 75% accuracy for faces of any ethnicity. When the software fails to detect a student's face, it flags them to instructors as potentially suspicious. Many students of color reported difficulties getting the software to recognize their faces, sometimes requiring extreme measures to be detected. The facial detection issues were discovered in 2020 during the COVID-19 pandemic when remote learning became widespread. The University of Illinois announced in February 2021 that it would discontinue using Proctorio by mid-2021, citing significant accessibility concerns, privacy violations, and discrimination issues. Over 1,000 students and staff signed petitions at various universities requesting discontinuation of the software. The software's discriminatory performance particularly affected marginalized students who were forced to use it for remote examinations during the pandemic.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Accuracy and effectiveness of AI decisions and actions are dependent on group membership, where decisions in AI system design and biased training data lead to unequal outcomes, reduced benefits, increased effort, and alienation of users.
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed