NUS uses AI to enhance ‘human skills’ among students
By Si Ying Thian
The National University of Singapore has been piloting GenAI-powered ScholAIstic since last November, and the school believes that AI can help scale roleplaying and the development of practical skills for some professions.

NUS has been piloting Gen-AI powered ScholAIstic LLM since last November for the purpose as roleplaying bots. Image: Canva
For domains like law, nursing and social work, large language models (LLMs) could potentially bridge the gap between academic instruction and real-world applications.
At the National University of Singapore (NUS), instructors are using LLMs like ScholAIstic to implement and scale roleplaying scenarios to teach students practical, human skills to meet the demands of their respective workplaces.

“Roleplaying with every student might not be scalable [for teachers]. But now with this LLM-based system, students could practice,” said NUS’ Associate Professor Ben Leong, who is also the AI Centre for Educational Technologies (AICET)’ Director and AI Singapore’s Chief Data Officer.
Prof Leong was presenting at the Adult Learning Xchange conference organised by the Institute for Adult Learning (IAL) on May 29 in Singapore.
During his presentation, he shared an example of ScholAIstic used by law students to learn courtroom cross-examination techniques.
What is unique about the LLM is that it could respond as different court participants, including the judge and opposing counsel. The student would act as the defence counsel to cross-examine the witness.
Additionally, the LLM would provide feedback on improper questioning by the student. For instance, if the student asked a leading question like “are you guilty?”, the judge would object and intervene.
With the growing use of artificial intelligence (AI) among students, how could educators strategically employ the technology to innovate and improve education?
To subscribe to the GovInsider bulletin, click here.
Pedagogy-first approach
Prof Leong said his centre prioritises understanding pedagogical challenges before determining how technology can solve them.
“The way forward is not about technology, but understanding pedagogy. You look at why students are not learning and how we can do better now. What can tech do for you?

Roleplaying bots, for instance, is one case where tech can do something that teachers cannot do themselves due to time constraints, according to Prof Leong.
“If we’re able to replicate something effective to tackle something manual, there is some hope to exploit tech to scale up," he added.
AICET was set up in December 2020 during the pandemic with the aim of better equipping educators to leverage tech for teaching, said Prof Leong.
Funded by AI Singapore, the office was jointly set up by the Smart Nation and Digital Government Office (SNDGO) and Ministry of Education (MOE). While it is hosted by NUS’ School of Computing, its scope of work extended beyond NUS.
Customising tech to teaching
The SchoALstic LLM was built to tackle a problem raised by Social Work professor Dr Gerald Chung at NUS to better prepare students for real-life challenges through guided roleplay.
To enhance the relevance and reliability of the model, the team used the retrieval augmented generation (RAG) technique to connect the LLM with a specified knowledge database.
Consultants at AICET would work with the professors from the respective courses to engineer and test the prompts used for the LLM.

Prof Leong highlighted two components to prompting the LLM.
The first was prompting a scenario alongside the case facts to facilitate the roleplaying process with the students.
The second prompt was to provide feedback on the student’s performance after the roleplaying exercise.
“The powerful thing is that we can see every interaction [between the LLM and the student],” he said, adding that the model was also able to effectively achieve both objectives of assessing and providing feedback.
GovInsider earlier did an interview with Prof Leong on some other ways that AI would be used in education, including exam grading and adaptive learning.
Prof Leong added that the nursing classes would start to actively use the model from this August onwards, while the law classes would start using it in the following semester coming next January.
“Hopefully this allows the students to practice and also try to improve the model to provide scenarios that are even more practical,” he said.
Editor's note (June 4): The writer clarified when the nursing and law classes, respectively, would start to actively use the ScholAIstic LLM.