> Rhetoric > Oracle-Mediated Rhetoric > arghXiv:2601.05529
Detection of Machine-Generated Text by Participants in Adversarial Exchanges
Abstract: Participants were asked to identify which of an opponent's messages had been produced by an automated system. Accuracy was 49.6 per cent against a chance rate of 50 per cent; mean reported confidence was 91 per cent. Accuracy did not improve when participants were told that half the messages were machine-generated, nor when they were told, truthfully, that none were.