‘Not about preventing AI misuse’: S’pore universities move from grading essays to assessing thinking

Over time, the detection tools have become obsolete, he elaborated, and research has shown that they produce unreliable results.

For example, SUSS had used AI detectors from Turnitin, a software used to check assignments for plagiarism.

The tool is based on statistical guesswork instead of concrete proof. According to research, it raises false flags for structured human writing, and unfairly targets non-native English speakers who write simply.

With NTU teaching students AI literacy and responsible use from April 2026, as well as offering them free access to a suite of premium Google AI tools, Wolfrum said that it was timely to retire AI-detection tools.

Other universities also offer students access to AI tools, such as ChatGPT Edu and Microsoft Copilot. Some use custom AI bots to assist with daily studying, coding and even oral exams.

“NTU’s approach to assessment is moving in a direction where faculty can evaluate what a student actually knows and understands, not just what they submitted,” said Wolfrum.

Grading understanding as students learn with AI

The local universities said their students are told explicitly when AI use is allowed, forbidden or is a must. For example, some courses require students to brainstorm with AI to develop a project, while others prohibit students from using AI when answering a quiz.

As more universities here phase out grading work submitted each term, they seek to evaluate the process of learning, or assess students’ thinking and understanding at progressive checkpoints.

Lee, SMU’s associate provost for undergraduate education, said the university implemented new assessment design guidelines for faculty to follow in the age of AI. One criterion mentioned is designing assessments where AI offers no meaningful advantage.

Across the six autonomous universities here, assessments come in various forms. These include:

In some instances, such as in SMU’s writing and reasoning course, instructors assign higher-order tasks where students are graded on their ability to evaluate AI outputs.

Students first use AI to generate summaries of readings. Then, they evaluate them for accuracy, bias and hallucinations by comparing the AI-generated material with published sources.

Designing such tasks takes time and involves rigorous testing before they are used. But SMU’s Lee said this is necessary and useful at a time when students have to learn with AI.

“We have no choice. We have to change how we are teaching and testing students to make sure they are learning,” she said.

Alvin Teo, 24, a third-year engineering student at SUTD, said while traditional exams or essay writing can be stressful for him, oral presentations require him to think fast on his feet and understand concepts in such depth that he can argue them convincingly in person.

Said Teo: “With oral defence, I cannot just memorise the final result. I have to be prepared for someone to ask, ‘Why did you choose this geometry? Why do you expect it to perform better?’

“If there is a gap in my understanding, I have to work through it and respond in real time.”

Xavier Goh, 21, a second-year economics undergraduate at SMU, said AI usage remains high among university students regardless of the assessment format.

“Shifting from written final exams to live presentations changes how we use AI rather than reduce our reliance on it,” he said.

While AI serves as an interactive tutor when he is preparing for traditional exams, Goh said its use for live presentations and oral examinations is limited to background research and script refinement. The technology cannot help him during in-person Q&A sessions.

Ashraf Kassim, SUTD’s associate provost of education and innovation, said assessment methods like oral presentations at different checkpoints keep students engaged throughout the course and help them develop a deeper understanding of the material.

“They can’t put off learning to the end and wait for the high-stakes exam paper,” he said, adding that this approach enhances students’ employability as it prepares them better for real-world problem-solving.

May Lim, SIT’s assistant provost for applied learning, said assessments should not focus only on whether AI was used, but also whether students can demonstrate the knowledge, competence and professional judgment expected of a graduate.

This, she said, is increasingly important in AI-enabled workplaces, where employers will expect graduates to combine disciplinary expertise with AI skills.

“Assessing is not just about preventing the misuse of AI. It is to ensure that AI does not replace the learning, thinking and capabilities that students are expected to develop,” said Lim.