The 87% AI-Detection Flag

'Clear and convincing evidence' is a demanding civil standard: it requires more than a bare probability, but less than beyond reasonable doubt. The 87% figure is not the probability that Sam cheated. It is the tool's output given its training data, and it must be combined with a base rate and a false-positive rate. Suppose 2% of essays are AI-generated and the tool has a 5% false-positive rate and 87% sensitivity. Then the probability that a flagged essay is AI-generated is about (0.87 × 0.02) / ((0.87 × 0.02) + (0.05 × 0.98)) ≈ 0.26, or 26%. That is far below clear and convincing evidence. Evidentialism asks the panel to proportion its belief to the total evidence, which includes Sam's six drafts, the teacher's observation about style, and the lack of local validation. The version history is counterevidence, though it could be fabricated. A reliabilist would ask whether the detection process is reliable in this context; it has not been validated on the school's students, and no expert is present. The panel member who says '87% is basically certain' confuses a likelihood with a posterior probability. The panel should not refer on this evidence alone.

Testimonial injustice occurs when a speaker's credibility is unfairly deflated because of prejudice or identity. Sam is a student, dependent on the school's judgement, and may be nervous; the panel could easily treat his denial as self-serving. To avoid this, the panel should treat Sam as a knower with a claim to be heard. It should ask him to explain his drafting process, identify specific sources, and answer questions about the argument. It should compare his oral account with the version history and the essay's content. It should not require him to prove his innocence, since the burden of proof is on the school. The panel should also be careful not to over-weight the teacher's impression of a 'sudden shift' in style; stylistic change can have many causes. A fair process would record Sam's account, give him time to respond, and separate the question of evidence from the question of punishment. The panel should recognise that its own credibility as an institution depends on not using an unvalidated tool to overrule a student's testimony.

A virtue epistemologist would focus less on the 87% and more on the character of the inquiry. The panel is being invited to display arrogance ('87% is basically certain'), haste, and closed-mindedness. The virtuous response is epistemic humility: the panel lacks technical expertise, so it should seek independent advice. It should be open-minded about alternative explanations for the stylistic shift. It should be careful and thorough: read the drafts, compare them, and ask Sam to explain his choices. It should be honest about the limits of the evidence and about the pressure from the parents' legal threat. A virtue approach would also ask what kind of school community the panel wants to be: one that treats students as potential knowers or one that treats them as suspects. The panel should probably postpone the decision, request an expert review, and document its uncertainties. This does not mean abandoning judgement; it means making the judgement in the right epistemic spirit.