A Chinese Reviewer Says AI-Written Theses Should Never Reach Outside Exam
A Chinese academic has made a blunt proposal that is circulating among researchers: dissertations written by artificial intelligence should not be sent out for external review at all. The argument, published as a personal essay on a Chinese science blog by a professor named Peng Bo, rests less on principle than on wasted hours. After several rounds of reviewing degree theses that struck him as fluent but strangely weightless, he concluded they were probably machine-written, and that only that explanation accounted for what he was reading.
The essay describes two patterns. In some cases an entire thesis appears to have come from a large language model, start to finish. In others, the student seems to have drafted parts independently and used AI to expand, smooth or pad the rest. Either way, the reviewer is left doing a job the system was not designed for: guessing authorship from prose style, then deciding whether a degree should proceed on that guess.
That is an uncomfortable position for Chinese graduate education, which relies heavily on external examiners. Doctoral and master’s theses are routinely sent to anonymous specialists at other institutions, who are expected to judge originality, method and command of the field. Those reviewers are usually unpaid or modestly compensated, and they work at scale. If a meaningful share of submissions is machine-assisted, the cost lands on exactly the people with the least time to absorb it.
The deeper difficulty is detection. Tools that claim to identify AI text have a poor record, producing false accusations against careful non-native writers and missing lightly edited machine output. Watermarking, which some model developers have begun to build into generated text, depends on the model and on nobody stripping it out. A reviewer holding a printed or PDF thesis has no practical way to check.
The professor’s suggestion, that such work simply not be sent out, raises its own questions. Someone has to make the call before the envelope leaves the department, and that someone is usually the student’s own supervisor, who has an interest in the thesis passing. Institutions would need a screening step with real authority, plus a route for appeal when a student is wrongly flagged.
Chinese universities have moved quickly to regulate AI in academic work, with many issuing rules that permit assisted editing while prohibiting wholesale generation. Enforcement, however, remains local and uneven. The essay’s real target may be less the technology than the incentive structure around it: students under pressure to produce length, supervisors stretched across too many candidates, and reviewers asked to certify quality they cannot verify.
For global readers, the story is a preview. Universities everywhere are wrestling with the same question, and most have answered it with honor codes and disclosure requirements rather than technical checks. What the Chinese case adds is a concrete proposal from inside the review system itself: stop sending the work out, and force the judgment upstream, where the supervisor and the department can be held to it.