AI writing detectors may wrongly flag some students, a review finds

A review of 30 studies says automated checks can mistake non-native English and structured academic writing for AI-generated text.

Editorial illustration — not from the study.

The review examined 30 empirical studies on false positives in systems designed to identify AI-generated writing. Across the studies, the researchers found that detectors often rely on measures of how predictable and varied text appears, which can lead them to classify some human-written academic work as synthetic.

When detectors flag writing

The review reports that current AI-writing detectors frequently misclassify structured academic discourse and non-native English writing as AI-generated. It identifies measures such as perplexity, which reflects how predictable a text is, and burstiness, which captures variation in sentence patterns, as important parts of these systems’ assessments. The synthesized evidence also points to the possibility that punitive enforcement could disproportionately affect specific linguistic groups, increase assessment anxiety and undermine fairness.

Why fair assessment matters

Universities use automated detectors as part of efforts to protect academic integrity, but the review describes exclusive reliance on them as a vulnerability in assessment quality assurance. The concern is especially relevant in English-medium higher education, where these tools are widely used and where differences in language background may affect how human writing is classified.

Evidence and caveats

This was a systematic review following PRISMA 2020 guidelines, with eligibility criteria based on the SPICE framework. The researchers searched Web of Science, Scopus and SciELO and conducted a qualitative, narrative synthesis of 30 primary empirical studies; the protocol was registered in OSF. The abstract does not provide a pooled false-positive rate, and the findings are presented as a synthesis of varied studies rather than a single controlled test of one detector.

Southern Democratic newspapers increased anti-Black propaganda when Populists threatened

Historical newspaper records suggest that Southern Democratic political elites used anti-Black propaganda more heavily when the Populist Party threatened their support among white voters. The increase was concentrated in Democratic-affiliated newspapers and was consistent with Democrats gaining politically in later elections.

18 hours ago·44 views

Society & Economics

Speech samples may track several kinds of fatigue in shift workers

In a study of 1,197 people across six countries, researchers used brief speech samples to estimate different aspects of fatigue. The recordings performed well for current sleep deprivation and sleepiness, while models tailored to each person were better at detecting physical and mental exhaustion.

23 hours ago·39 views

Society & Economics

20 children’s mobile games all used deceptive design, study finds

A study of 20 popular free children’s mobile games in German-speaking regions found deceptive design in every app. Advertising appeared in 90% of the games and was usually non-skippable and woven into gameplay.

1 day ago·48 views