Probability is not proof: Why AI detection differs from plagiarism detection in academic misconduct cases
TL;DR
It is concluded that detection outputs should never serve as standalone proof and recommends process-based assessment, bias audits, and transparent policies.
Abstract
This conceptual synthesis examines why artificial intelligence (AI) text detection cannot be treated as equivalent to plagiarism detection in cases of academic misconduct. Plagiarism detection tools operate forensically: they identify specific passages, match them to verifiable sources, and produce evidence that both instructors and students can examine and rebut. AI detectors operate statistically: they estimate the probability that a text resembles AI-generated writing, without identifying a source, act, or comparator. Drawing on independent evaluations showing no tool exceeds 80 per cent accuracy,1 false-positive rates above 61 per cent for non-native English writers in a Test of English as a Foreign Language (TOEFL)-essay testing condition,2 and journalistic survey evidence reporting disproportionate false-accusation experiences among Black students,3 the paper applies three lenses. First, evidentiary: AI scores are not falsifiable and suffer base-rate problems that make predictive value unknowable in real classrooms. Secondly, procedural: under Goss v. Lopez (1975) and Mathews v. Eldridge (1976), students are entitled to notice and a meaningful opportunity to respond, a safeguard undermined when the evidence is an opaque probability. Thirdly, fairness: applying Rawlsian justice, rational agents behind a veil of ignorance would reject a system whose errors fall most heavily on non-native speakers and on groups that survey evidence suggests may face disproportionate risk of accusation. Recent reporting on one federal preliminary ruling suggests discipline is more defensible when institutions rely on corroborating evidence, not scores alone. The paper concludes that detection outputs should never serve as standalone proof and recommends process-based assessment, bias audits, and transparent policies. This article is also included in The Business & Management Collection which can be accessed at https://hstalks.com/business/.