Lecturers at four Singapore universities use AI to grade students’ work

“In this instance, AI acts as a supporting partner to the instructor,” said SUTD’s associate provost of education and innovation Ashraf Kassim.

At SIT, an internally developed platform called AI-Orate created a chatbot to quiz food technology degree undergraduates on their understanding of food manufacturing processes and grade them based on their answers. Chatbots like this one can mimic a human examiner by asking follow-up questions to drill down on concepts.

About 50 students took part in one of the few runs of AI-Orate in October 2025, during which they wrote code to program an industrial machine to carry out heat processing to predict microbial reaction rates at different temperatures. They were then quizzed by the chatbot on their reasoning, with follow-up questions to drill down on their responses. The chatbot then generated a transcript of the conversation along with a recommended grade.

Associate Professor Wong Shin Yee, programme leader for SIT’s Bachelor of Food Technology with Honours programme, said that the chatbot can cut down time-consuming face-to-face assessments from one week to about two days.

“There is a lot of potential, especially for creating highly customisable assessments in large classes where personalised questioning is usually impossible,” said Prof Wong, adding that chatbots can be applied to group submissions to evaluate how well each student understands their group project.

Associate Professor Karin Avnit, deputy director at SIT’s Teaching and Learning Academy, said: “The adaptive nature of the discussion allows students to demonstrate their competence.”

NUS permits AI use only if the tool has been validated for accuracy and approved by the relevant head of department or deanery, under a university forum that sets AI application standards.

Only one tool used for two post-admission English tests meets this stringent requirement. The tests are for students lacking the necessary English qualifications.

The tool, which has been in use since July 2025, assesses argumentative essays on content, organisation and language. It has been trained to emulate the standards of an expert panel of human NUS graders.

To ensure reliability, AI performs the grading twice, and human graders are also roped in to audit the AI’s results and examine borderline cases.

“We have observed that this hybrid approach significantly enhances grading consistency, compared with solely relying on human graders who can be susceptible to fatigue,” said NUS’ associate provost for education and technology Melvin Yap.

SMU and SUSS are the only universities that have yet to approve AI in grading scripts that count towards a student’s final result.

SUSS said it prefers to rely on humans for consistency for now.

“Until there is a more robust understanding of the accuracy and reliability of AI in different grading contexts, human academic judgment must remain central in high-stakes assessments,” said SMU’s vice-provost in charge of education Venky Shankararaman.

The university started a working group to evaluate the effectiveness and limitations of AI-enabled grading tools in mid-2025.

An SUSS spokesman said the varsity prefers to use humans to enforce established marking rubrics. “This ensures that grading remains consistent, transparent and aligned with academic standards.”

NUS undergraduate Leslie De Souza, 23, who majors in English literature, said he does not trust an automated tool to understand the nuances of his work. “How would AI tell me what to improve on?”

NTU undergraduate Ryanna Lee, 22, who majors in communications studies, hopes for more checks and balances. “It’s only fair that AI use by teachers is watched and controlled,” she said. 

However, NUS undergraduate Kwan Kaiee, 22, who is taking a double degree in history and English literature, said the use of AI is inevitable. “University lecturers and professors in Singapore are overworked due to their dual commitments of research and teaching.”

SIT undergraduate Lief Chng, 26, who took part in the October chatbot examiner trial, said he and his classmates were initially sceptical about how well the chatbot would be able to capture the nuances of their replies and assess them. After the test, Mr Chng was more open to the idea of a chatbot examiner.

“In a normal test, you either get it right or wrong. But in the chatbot, it feels like you have a second chance to earn marks because the chatbot can prompt us further if our first answer is not complete or good enough,” he said.