2.2 Evaluation of Automated Detection Reliability and Algorithmic Vulnerabilities
The operational tension between automated artificial intelligence detection software and authentic assessment models reflects a deeper divergence in institutional philosophy across higher education [2]. Automated detection systems are frequently deployed under the premise of maintaining procedural compliance and deterring non-original authorship [1]. However, academic governance documentation and pedagogical scholarship demonstrate that automated classifiers are inherently susceptible to algorithmic ambiguity, yielding significant rates of false positive classifications that disproportionately penalise non-native language learners and erode institutional trust [2]. Furthermore, reliance on automated surveillance mechanisms reduces academic integrity to a punitive transaction, failing to encourage meaningful intellectual engagement or critical reflection [7]. In contrast, authentic assessment restructures evaluative tasks around contextual problem-solving, iterative drafting, oral defence, and applied domain performance, thereby making the assessment intrinsically resilient against unreflective automated synthesis [1], [7]. Despite these pedagogical benefits, the broader implementation of authentic assessment within Canadian universities faces structural impediments, particularly regarding faculty workload, inadequate technical training, and fragmented policy environments [1]. When authentic evaluation is treated solely as an informal, discretionary exercise left to individual instructors, its systemic efficacy remains compromised [1]. Resolving this crisis requires Canadian universities to move beyond procedural surveillance tools and actively establish organizational infrastructure that funds, trains, and standardises authentic assessment practices across all academic faculties [1], [2].