Predicting Dropout Risk, Breaking Privacy Law — Quebec's Val-des-Cerfs School Board AI Ruling

Starting around 2017, the Centre de services scolaire du Val-des-Cerfs in Quebec's Eastern Townships worked with data specialists at the accounting firm Raymond Chabot Grant Thornton to build a machine-learning model predicting which Grade 6 students were at significant risk of dropping out three years later. The model drew on more than 300 data points collected since 2002 on 60,000 students, including academic results, financial-aid status, absenteeism, disciplinary history and frequent address changes, with identifying fields such as names and addresses stripped before analysis.

CBC News reported in November 2018 that the model reached 92% accuracy on its own retrospective test, and that at the end of the 2017–18 school year it had flagged about 90 incoming Grade 7 students as at risk, with the contributing factors sent to their schools. Quebec's Ministry of Education subsequently asked a provincial research centre to study extending the approach system-wide.

Following media coverage, Quebec's data-protection authority, the Commission d'accès à l'information (CAI), investigated and ruled in December 2022 that the school board had violated provincial privacy law: the stripped data was de-identified but not properly anonymized, meaning students remained re-identifiable through other institutional records; the algorithm's risk scores counted as newly created personal information; and parents and students were never informed their data was being used this way. The CAI ordered the board to notify affected families of the project's purpose, data sources and their rights. The case is a documented cautionary example of an AI system with reported predictive strength but no lawful basis or transparency in how it processed sensitive student data.

Read the full analysis: https://www.cbc.ca/news/canada/montreal/school-board-measures-dropping-out-risk-1.4887141