何が起こったのか
The Straits Times reports that Singapore universities are redesigning assessments as AI use becomes more common. Institutions including NTU, SUSS, NUS, SIT, SUTD and SMU are using or considering live presentations, oral defences, in-class writing, staged submissions and assignments that require students to evaluate AI-generated material.
The Straits Times reports that Singapore Management University is using live assessment to test students’ understanding in ways that are difficult to complete through a prepared or AI-generated submission alone. In one management communication module, a student’s five-minute presentation and follow-up questions account for 30 per cent of the final grade. The report says the professor evaluates whether students understand their concepts and can respond spontaneously, alongside weekly reflections and essays completed in class or through a lockdown browser.
The newspaper also reports that SMU students in a sustainable marketing class were given an essay question two weeks in advance and allowed to use AI for research and answer development, but then had 15 minutes to write the essay in class without reference materials. According to the report, the design is intended to show whether students can turn their research and understanding into a coherent answer rather than use AI as a shortcut. Other reported methods include gallery presentations, reflective journals completed in class, oral defences and staged submissions that show how a project develops.
The Straits Times says the change is occurring alongside a retreat from AI-detection tools. NTU and SUSS said in August that they would stop deploying such tools to flag suspected AI misuse, while NUS, SIT and SUTD told the newspaper they do not use them. SMU said detection tools are only one way it identifies misuse. NTU’s deputy president and provost, Christian Wolfrum, told the newspaper that detectors once served a reasonable purpose after ChatGPT’s release but had become obsolete as research showed unreliable results. The report says SUSS had used Turnitin’s AI detector and describes the tool as relying on statistical rather than conclusive proof.
The report says universities are not adopting one universal rule for AI. Students are told when AI is allowed, forbidden or required, depending on the course. In some classes, they may use it to brainstorm or research; in others, they may be prohibited from using it in a quiz. SMU’s reported writing-and-reasoning exercise requires students to generate summaries with AI and then assess those summaries for accuracy, and hallucinations by comparing them with published sources. The Straits Times reports that designing and testing such assignments requires substantial work from instructors. The universities’ policies, the scale of implementation and the results of these methods were not independently confirmed in the source provided.
なぜそれが重要なのか
The shift changes the educational question from whether a student used AI to whether the student can demonstrate knowledge, judgment and reasoning. It also reflects growing concern that AI detectors can produce false positives and unfairly affect some forms of writing, although the claims and institutional practices described here have not been independently confirmed.
The reported change treats assessment as a way to observe learning rather than merely inspect a final artifact. A polished essay can show writing ability, but by itself may reveal little about how a student selected evidence, made decisions or understands the underlying ideas. Live questioning, drafts and in-class writing give instructors additional evidence about the reasoning process. That does not prove that the student worked without AI at every earlier stage, but it can make unsupported understanding harder to conceal during the assessment itself.
The move also addresses a practical weakness in automated AI detection. The Straits Times reports that such systems can produce false flags for structured human writing and may unfairly target non-native English speakers who write in simpler forms. If an institution treats a probabilistic detector as proof of misconduct, students could face penalties without a reliable basis for the decision. Retiring or limiting those tools may reduce that risk, but the source does not independently establish the accuracy of the research cited or explain how each university will investigate suspected misuse instead.
For students, the change means that AI literacy becomes part of academic competence rather than a question confined to discipline. They may need to know how to use an AI system for research, recognize hallucinations and , document their process, and defend conclusions without the system present. The Straits Times quotes students who say oral assessments demand deeper preparation and real-time reasoning, while another student says AI use remains high but shifts toward background research and script refinement. The implication is not that live assessments eliminate AI dependence; they change where and how that dependence is visible.
For employers and the public, the stakes extend beyond university grading. The newspaper reports that SIT sees graduate assessment as a test of knowledge, competence and professional judgment in AI-enabled workplaces. Graduates who can produce an AI-assisted answer but cannot explain or challenge it may be less prepared for roles where errors carry financial, legal, health or safety consequences. At the same time, the source provides no evidence that the new formats improve employment outcomes, learning retention or fairness across different student groups. Those are open questions rather than established results.
インタラクティブなメカニズム: 実際にどのように機能するか
この開発の背後にある基盤となるテクノロジーをインタラクティブに探索します。
Why can ethical evaluation not be reduced to one model score?
次に見るべきもの
The Straits Times reports that universities are clarifying when AI use is allowed, prohibited or required. The important next test will be whether these assessment formats improve learning without creating excessive workloads, disadvantaging students with communication difficulties or simply moving AI use into less visible parts of coursework.
The first issue to watch is consistency. The Straits Times reports that six autonomous universities use a range of assessment formats and that individual courses decide whether AI is allowed, forbidden or required. That flexibility may fit different disciplines, but it can also create uncertainty when students move between classes with different rules. Clear instructions about permitted tools, disclosure, data handling, citation and sanctions will matter as much as the format of the assessment itself.
The second issue is workload and access. Live presentations, repeated checkpoints and oral defences require time from instructors and students, especially when classes are large. They may also place additional pressure on students with disabilities, anxiety, language differences or limited access to preparation support. The source reports that some students find oral assessment demanding, but it does not provide comparative data on outcomes or accommodations. Universities will need to show whether the new designs are workable and equitable at scale.
A third question is whether assessment can keep pace with AI-assisted preparation. A student may use AI to research, brainstorm, draft or rehearse before an in-person evaluation, even when the final performance is unaided. That is not necessarily misconduct if the course permits it, but it makes transparency important. The reported assignments that ask students to critique AI output offer one possible approach: they test whether students can use the technology while checking its accuracy and limits. The source does not say how often these tasks are used or whether they have been validated against independent measures of learning.
Finally, universities will need evidence about replacement safeguards after detector tools are phased out. The Straits Times reports that NTU is teaching AI literacy and responsible use from April 2026 and offering students premium Google AI tools, while other institutions provide access to ChatGPT Edu or Microsoft Copilot and use custom bots for activities such as studying, coding and oral exams. It remains unknown how institutions will audit these systems, protect student data, handle disputed cases or measure whether students are actually learning. The strongest test of the reported shift will be transparent results on learning, fairness, workload and academic-integrity decisions over time.