Students are increasingly benefitting from personalized coaching by a large-language-model teaching assistant created by Tsinghua university.
Every fall semester heralds a fresh start, but September 2023 marked a particularly exciting beginning at Tsinghua University: for the first time, artificial intelligence (AI)-powered study aids, known as ‘AI teaching assistants’ were introduced to students.
The multi-modal agents were incorporated into about 117 courses, spanning 40 departments and 10 fields of study, including thermodynamics, writing and communication, environmental policy, and more. To students, the AI teaching assistants provide 24-hour personalized learning support and guidance via a platform accessible from an app and online portal.
The AI teaching assistant works in a similar fashion to other large language model-based systems, such as ChatGPT or Ernie Bot: students type a query into the chat box and quickly receive written replies on everything from coursework to suggested knowledge extension exercises, explains Wang Hongning, who led the program’s technical development through Tsinghua’s Department of Computer Science and Technology.
“The answers given by the AI are derived from large, course-specific knowledge bases composed of textbooks, previous class materials, technical reports, and research papers compiled by the lecturers,” explains Wang.
Students can have several different types of interactions with the AI teaching assistants, from simple, conversation-style question-and-answer sessions, to step-by-step guidance, feedback on writing assignments, and quizzes that encourage more extensive exploration of a topic.
Scroll down to see our list of the five best things about the AI teaching assistants.
The team spent more than a year refining the large language model, including tackling bugs and addressing concerns such as bias and data privacy. For example, they made sure the AI teaching assistants provide sources for all the information they generated, to combat the factual inaccuracies that plague the results from some large language model tools.
They also worked with subject specialists to improve the model’s accuracy, and consulted pedagogical experts from Tsinghua’s Institute of Education to figure out how to enhance the students’ learning experience and engage their critical thinking skills.
Wang says it was the explosion of large language models globally that sparked the impetus for the new AI teaching assistant. “Here are systems that can respond to questions, clarify concepts, and generate narratives and solutions — activities at the very heart of education,” Wang says. “We wanted to understand how they could enhance the Tsinghua’s capabilities, and to explore the broader role of generative AI in the learning environment.”
While a few teachers were initially hesitant about the new initiative, many have now jumped on board, he says. “They’ve seen their colleagues using it successfully and are eager to try it themselves.” But the true satisfaction for Wang comes from witnessing the impact on students: “I’ve watched them grow more confident through their interactions with the AI, and that’s what I’m most proud of.”
Here are five ways AI teaching assistants are making a difference at Tsinghua:
1. Helping students tailor their learning
Initial testing indicates that a group of students who received feedback from AI teaching assistants improved their grades by 10% on a specific assignment in the course. However, Wang is quick to point out that the goal is not simply to improve grades, but to create a more self-directed academic trajectory for students, who can use the AI teaching assistants to help broaden their learning scope and to facilitate bespoke learning experiences based on interests and aspirations.
Personalized teaching assistants have the potential to change the nature of higher education, says Wang. He believes the future of higher education will see students able to navigate even more specialized and personalized pathways through their degrees.
2. Creating a safe space to ask questions
One of the perks of having an AI teaching assistant is that it makes it less intimidating for students to clarify doubts. “During lectures, discussion time is limited, and some students might be too shy to ask questions,” says Long Ying, a professor at the School of Architecture who introduced AI into his urban planning course, The New Science of Cities, in 2023. “Others may not have the time to meet with teachers or teaching assistants, and some simply wouldn’t ask because they don’t yet understand what to ask.”
Tsinghua graduate student, Dai Xusheng, first encountered the AI teaching assistants as part of his Master of Education program. He agrees that the format can help reserved or overwhelmed students. “Imagine sitting in a lecture hall with 200 other students,” he says. “Even if you have a question, you’re hesitant to speak up because it feels rude to interrupt, and you worry about looking uninformed—especially in such a competitive place as Tsinghua.”
But with the AI teaching assistant, “you no longer feel pressured by others’ perceptions,” says Dai. “It’s a very safe environment that encourages me to ask questions.”
The platform is also available 24/7, although statistics reveal that usage peaks during class times and in the evenings. Human instructors try to provide such attention, says Wang, but with large class sizes, limited time, and a heavy workload, it’s often exhausting to provide detailed feedback to students [on every one of their assignments]”.
A lack of detailed feedback can be discouraging, says Wang. “AI can provide feedback so that students feel like their efforts are worthwhile, and that motivates them to keep improving,” he says.
That’s not to say human lecturers aren’t important, says Wang. These AI teaching assistants will just free them up to work on mentoring students at crucial junctions in their academic careers, to give extensive feedback on key pieces of work, or to put together more creative and cutting-edge aspects of the courses (see point 4. ‘Freeing up instructors’ time’), he says.
3. Training students to think more deeply, in a personalized way
While developing and refining the system, Wang and his colleagues placed great emphasis on creating a teaching assistant that wouldn’t just spoon-feed students the answers they sought. Rather, they wanted to guide students to figure out solutions for themselves. “The education process isn’t about memorization,” explains Wang. “It’s about inspiring students and teaching them how to think.”
His team designed the platform so that when a student types in a question, the AI teaching assistant sometimes counters by asking related questions, or suggests that the student come up with a solution that the student and AI can then discuss further.
At the same time, the system analyzes why a student might have posed such a question, gauges their level of understanding, and identifies gaps in their knowledge. This information helps guide further AI interactions and conversations with the student, with the AI adjusting to the student’s level and needs accordingly. “We are trying to make the experience more contextualized and personalized,” Wang says.
4. Freeing up instructors’ time
Another core function of the AI teaching assistants is to act as virtual teaching assistants to teachers, meaning the instructors can ask for help planning their curriculum, obtaining teaching materials, and mapping class activities. But the real game-changer has been with respect to time, says Long. “Having AI teaching assistants help students with some of the baseline knowledge allowed me to focus on preparing deeper, more comprehensive and cutting-edge materials. It leaves more time for creativity.”
He believes that AI teaching assistants are especially useful for general courses, in which broad cross-disciplinary concepts are taught. “Since students on such courses come from various disciplines, the AI teaching assistant can help them quickly get up to speed on the material,” says Long. “Otherwise, these students often need multiple discussions with human teaching assistants or instructors offline, or during limited class hours, to grasp the content.”
5. Making education more equitable
Sophisticated large language models designed to aid students will help make education more accessible, says Dai, who plans to pursue a career in education technology. “There are many remote areas in China that face a lack of resources and qualified teachers. Technology, especially the emergence of generative AI, is a great way to resolve this issue and make a change.”
Geographical limitations aside, AI teaching assistants will also improve educational access for those unable to attend in-person sessions due to work commitments, or health and mobility issues. “We really want to push this effort outside Tsinghua,” says Wang, “because then we’ll see the impact of our system in bridging inequality in educational opportunities in a broader context.”
