Publishing industry warns of a “witch hunt” over AI text detectors
A La Vanguardia report describes accusations of AI use based on detectors whose verdicts contradict themselves over time. The debate reaches education: how to check authorship without hunting for culprits.
A report published on 1 November in La Vanguardia describes how publishers and literary agencies are starting to run manuscripts through AI detectors such as Pangram. Several people in the sector speak of “paranoia” and a “witch hunt”: authors who say they have not used AI are accused and do not know how to prove it, and advice is circulating to keep a text’s version history in case authorship has to be demonstrated. The report also cites a recent case involving a novel awarded a prize by a Mexican publisher and points out that these tools “are not infallible”: two novels from 1969 and 1946 that showed traces of AI in earlier tests now come out as human with the same detector, according to the newspaper’s own check.
In the group, the conversation moved the issue to student assessment. One view holds that the problem is blanket suspicion: the detector is sometimes right and sometimes wrong, so checking should rely on knowing the person, on triangulating authorship with other formats and contexts on the same subject, and on judging whether the signatory could have written what they submit, as the report itself suggests when it proposes a conversation with the author.
Another view insists that, before monitoring, a framework must be set: whether AI is allowed, for what uses, with which detector and metrics, and what evidence an accused person can provide. Without clear rules, it warns, suspicion has no limit; it also expresses confidence that, over time, AI and human creation will each find their own space. There was agreement that the permitted uses of AI and their purpose need to be clarified.
Publishing industry warns of a “witch hunt” over AI text detectors
A La Vanguardia report describes accusations of AI use based on detectors whose verdicts contradict themselves over time. The debate reaches education: how to check authorship without hunting for culprits.
A report published on 1 November in La Vanguardia describes how publishers and literary agencies are starting to run manuscripts through AI detectors such as Pangram. Several people in the sector speak of “paranoia” and a “witch hunt”: authors who say they have not used AI are accused and do not know how to prove it, and advice is circulating to keep a text’s version history in case authorship has to be demonstrated. The report also cites a recent case involving a novel awarded a prize by a Mexican publisher and points out that these tools “are not infallible”: two novels from 1969 and 1946 that showed traces of AI in earlier tests now come out as human with the same detector, according to the newspaper’s own check.
In the group, the conversation moved the issue to student assessment. One view holds that the problem is blanket suspicion: the detector is sometimes right and sometimes wrong, so checking should rely on knowing the person, on triangulating authorship with other formats and contexts on the same subject, and on judging whether the signatory could have written what they submit, as the report itself suggests when it proposes a conversation with the author.
Another view insists that, before monitoring, a framework must be set: whether AI is allowed, for what uses, with which detector and metrics, and what evidence an accused person can provide. Without clear rules, it warns, suspicion has no limit; it also expresses confidence that, over time, AI and human creation will each find their own space. There was agreement that the permitted uses of AI and their purpose need to be clarified.
