The comparison
Detection vs verification
The two approaches answer different questions. A detector asks: does this text look machine-written? Verification asks the question teachers actually care about: does this student understand the work they handed in?
AI detector
Reads: the text alone · Outputs: a probability score · Evidence: none · Student's recourse: none · Known failure modes: false positives, bias against non-native writers, paraphrase evasion
Viva oral assessment
Reads: the student, live, about their own submission · Outputs: transcript, understanding score, authenticity verdict · Evidence: the student's own spoken explanation · Student's recourse: the conversation itself is their defense · Teacher makes every final call
Viva doesn't accuse anyone. After every submission, each student has a short voice conversation about their own work — what they argued, why they chose their method, what they'd do differently. A student who did the work finds this easy, even enjoyable. A student who didn't finds there is no script for follow-up questions about choices they never made. Either way, the teacher reads evidence, not a percentage.
Some schools keep a detector as a weak preliminary signal, and that's fine — but when a grade, a scholarship, or an integrity record is on the line, a probability isn't a basis for a decision. A conversation is.