Singapore – Marking technical work on math or physics paper scripts is a painstaking task.
Recently, lecturers at Nanyang Technological University (NTU) and Singapore University of Technology (SUTD) have turned their attention to an artificial intelligence tool called Gradescope.
The tool scans handwritten responses and groups similar responses together, allowing human examiners to assign the same grade to all students who perform similar actions. Batch assignment saves time by eliminating the need for humans to comb through every script.
NTU and SUTD are among the four public universities that allow the use of AI tools to grade student work that contributes to the final score. The remaining two are National University of Singapore (NUS) and Singapore Institute of Technology (SIT).
Singapore Management University (SMU) and Singapore University of Social Sciences (SUSS) do not allow the use of AI to score student work towards end-of-year grades, with accuracy and reliability being the main concerns.
Broad debate surfaces over the role of AI in higher education in 2025 when NTU imposed penalties on three students For using AI in assignments. All six universities here are generally Allowing students to use generative AI to varying degreesacademic integrity requires declaring when and how such tools will be used.
Many discussions are centered around AI, but hinders student learning and critical thinkinglittle is said about unfair or inaccurate scoring by AI.
NTU, SUTD, NUS and SIT told The Straits Times that all AI-generated grades must be reviewed by teachers. Students must also be notified if AI is used for assessments that count towards their final grade for the year. Students can also request a reconsideration of their results.
Starting in August 2024, NTU will allow lecturers to use AI to mark mid-term and final exams for some physics and mathematics modules.
“This approach improves the consistency and efficiency of grading and puts all grading decisions in the hands of instructors,” said Christian Wolfram, NTU Vice-Chancellor and President.
SUTD began using Gradescope in April 2025 in Computer Science and Design. Schools evaluate tests that combine short-answer and explanatory questions.
“In this case, AI acts as a support partner for the instructor,” said Ashraf Qasim, vice president for education and innovation at SUTD.
At SIT, an in-house developed platform called AI-Orate has created a chatbot that quizzes undergraduate food technology degree students on their understanding of food manufacturing processes and scores them based on their answers. Such chatbots can imitate human examiners by asking follow-up questions to probe concepts.
In October 2025, approximately 50 students participated in one of the few executions of AI-Orate, in which they wrote code to program an industrial machine to perform heat treatments to predict microbial reaction rates at different temperatures. They were then questioned by the chatbot about their reasoning and asked follow-up questions to dig deeper into their answers. The chatbot then generated a recording of the conversation along with a recommended grade.
Associate Professor Wong Sing Yee, program leader of SIT’s Honors Bachelor of Food Technology programme, said chatbots can reduce time-consuming face-to-face assessments from one week to about two days.
“There are many possibilities, especially when it comes to creating highly customizable assessments in large classes where individualized questions are usually not possible,” Professor Wong said.added that chatbots can be applied to group submissions to assess how well each student understands the group’s project..
Associate Professor Karin Avnit, Deputy Director of SIT’s Teaching and Learning Academy, said: “The adaptability of the discussion allows students to demonstrate their abilities.”
NUS only allows the use of AI if the accuracy of the tool has been verified. Approved by the relevant department chair or dean, based on the University Forum that sets standards for AI applications.
Only one tool is used for the two post-admission English tests. This strict requirement. This test is intended for students who do not have the required English language qualifications.
toolin use since July 2025. evaluate Argumentative essays on content, structure, and language. Trained to emulate the standards of an expert panel of human NUS graders.
To ensure reliability, the AI performs scoring twice. Human scorers also collaborate to audit the AI’s results and investigate borderline cases.
“We found that this hybrid approach significantly improved the consistency of marking compared to relying solely on human graders, who are prone to fatigue,” said Melvin Yap, NUS vice-chancellor for education and technology.
SMU and SUSS are the only universities that have not yet approved AI for marking scripts that count toward students’ final grades.
SUSS said it currently prefers to rely on humans to ensure consistency.
“Until we have a more solid understanding of the accuracy and reliability of AI in a variety of scoring situations, human academic judgment must remain at the center of high-stakes assessments,” said Venky Shankararaman, vice president for education at SMU.
The university has set up a working group to evaluate the effectiveness and limitations of AI-enabled scoring tools in mid-2025.
A SUSS spokesperson said universities prefer to use humans to enforce established grading rubrics. “This ensures that grading is consistent, transparent and consistent with academic standards.”
NUS undergraduate Leslie de Souza, 23, majoring in English literature, said she did not trust automated tools to understand the nuances of her research. “How can AI tell us what to improve?”
Lyanna Lee, 22, an undergraduate communication studies student at NTU, would like to see more checks and balances. “It makes sense that teachers’ use of AI would be monitored and controlled,” she says.
But NUS undergraduate Kwan Kaie, 22, who is pursuing a double degree in history and English literature, said the use of AI was inevitable. “Singapore’s university lecturers and professors are overworked due to their dual responsibilities of research and teaching.”
Leaf Zhang, 26, an undergraduate student at SIT who took the chatbot examiner exam in October, said he and his classmates were initially skeptical about how well chatbots could capture and evaluate the nuances of replies. After testing, Chng became more open to the idea of a chatbot examiner.
“In a normal test, you either get it right or you get it wrong. But with a chatbot, it feels like you have a second chance to score because the chatbot can prompt you further if your first answer is incomplete or not good enough,” he said.
-
Additional reporting by Letitia Chen
