第18期:新加坡大学叫停AI检测工具 | 「用英语看世界」

Level Read

2026.10.17 第18期

    Level 1    

A university in Singapore has stopped using an AI writing detector across the school. The Singapore University of Social Sciences, or SUSS, turned off Turnitin's AI writing score on Aug 11. The school said the score has limits. It cannot conclusively determine if a student used AI wrongly.

Other universities in Singapore are also careful. NUS, SUTD and SIT do not use automatic AI detectors. They say a score alone cannot prove academic misconduct. Students are still responsible for honest work. When AI is allowed, they must follow rules and explain their use.

Teachers may instead ask for drafts, reflections, or a student's explanation. The important question is whether a student understands and owns the work they hand in.

    Level 2    

Singapore University of Social Sciences has disabled Turnitin's university-wide AI writing score from Aug 11. SUSS said such scores have limitations: they cannot conclusively show that generative AI was used inappropriately. The change does not mean students may ignore academic integrity. They remain responsible for the accuracy and honesty of their submissions.

NUS, SUTD and SIT told CNA that they do not use automated AI detectors. SIT had joined an early pilot, but did not adopt the tool because of its limits. These systems use statistical models to estimate whether text may have been generated by AI. A probability score is not proof that misconduct happened.

SMU still uses AI detection tools, but says they are one input among several, never the sole basis for a finding. Universities may look at the assessment context, a student's explanation, drafts and reflections. They are increasingly designing assessments that let students demonstrate real understanding, rather than trying only to detect AI after the work is submitted.

    Level 3    

The Singapore University of Social Sciences disabled Turnitin's university-wide AI writing score from Aug 11. Its reason was not that AI use no longer matters. SUSS said a detection score has limitations and cannot conclusively establish whether generative AI was used inappropriately. It still expects students to protect academic integrity and use AI responsibly.

The wider picture is similar. NUS, SUTD and SIT said they do not use automated AI detectors to decide plagiarism or misconduct. SIT said these tools use statistical models to estimate the likelihood of AI-generated content; probability scores alone cannot conclusively determine wrongdoing. NUS said present technologies have significant limits in reliability and accuracy, so they are not disciplinary evidence.

SMU keeps detection tools as one input among several, never the sole basis for a decision. Schools may consider assessment context, explanations, drafts and reflections. At SUSS, permitted support can include brainstorming, developing ideas, drafting and editing. Students must still acknowledge allowed AI use and show that they achieved learning outcomes.

This changes the educational question. Instead of asking only whether software can catch AI, teachers may design work that reveals understanding and ownership. Oral discussion, process records and knowledge of a student's earlier work can offer useful evidence. But these methods require careful, fair judgement; no single score should decide a student's case.