At right, Jiheng Jing discusses AristAI during a Teach, Talk, Listen, Learn podcast recording.

University of Illinois Ph.D. Student Leads Development Project of an AI Teaching Assistant

By Robert Baird, CITL Senior Associate Director

Jiheng Jing is a frequent and skilled user of GenAI, using it for his projects, studies, and recent entrepreneurial efforts. As a TA in Mechanical Engineering, Jiheng has had a firsthand view of GenAI use among students and instructors. Using GenAI daily for personal, professional and academic uses enabled Jiheng to move quickly and co-found AristAI, a startup dedicated to developing AI teaching assistants. These AI teaching assistants are designed for university courses, offering 24/7 responses to students' course-related inquiries, and guiding them to the relevant course content for review.

According to Jiheng: “The main idea is that we use all the course materials to train this AI chatbot [to answer] logistic questions, grading questions or homework questions . . . but also redirect students to the exact location of those lecture materials. If it’s a lecture recording, we will take students to the exact second of that lecture recording.” Additionally, the AI teaching assistant generates detailed analytical reports for instructors, offering valuable insights to help instructors refine and improve teaching strategies.

Jiheng met AristAI’s faculty advisor Dr. Tony Zhang, (Gies College of Business Adjunct Clinical Assistant Professor) through the 2022 Cozad New Ventures Challenge. Dr. Zhang had begun exploring AI when he became a Gies College Disruption Lab Fellow in 2021, originally focusing on how AI could assist with student learning around cryptocurrencies. Later, and months before the release of ChatGPT, Dr. Zhang pivoted his project and began exploring how AI could help develop virtual teaching assistants when he realized their potential for helping very busy instructors. While AI developers can create custom chatbots, Dr. Zhang knew that most instructors lacked the time and skills to develop their own virtual TAs.

The AristAI team, comprised of ten students and recent graduates from Illinois, Michigan, Pennsylvania, UCLA and Harvard, is actively involved in both the technological development and the marketing strategy for the AI teaching assistant. Jiheng serves as a leader of the AristAI team, which Dr. Zhang values for Jiheng’s collaborative skills and “success in leadership and willingness to take risks.” In the future, Dr. Zhang believes students and instructors will “leverage the GenAI more often and efficiently to provide customized learning curriculum and customized engagement with course instructors.”

Jiheng knows Illinois well, having started as a freshman in 2015 and is currently finishing his last year as a Ph.D. candidate. As a graduate student aligned with Dr. Gaurav Bahl's Mechanical Science and Engineering research group, Jiheng was well-positioned to understand the value of new technologies and the career paths available for engineering students. His expertise and appreciation for GenAI stems from his extensive personal and academic use of the tools, where he has been able to refine his language skills, develop computer code, and collaborate with others in the quickly developing world of GenAI.

As a teaching assistant, Jiheng witnessed some of the flaws in the traditional office hours and lab section formats: students asking the same simple questions over and over; students needing to be led to information they already possessed; and some students never seeking help. As an international student, Jiheng was personally aware of the challenges for those speaking English as a second language: difficulty of English grammar, the nuances and colloquialisms of English, as well as the complex vocabularies in academia and in STEM fields. Jiheng explained that when he first came to Illinois he was afraid to ask questions. “I’m afraid that the TA will judge me, even though I know that they won’t. And because I was a TA last year, I never judged students. But as a student, especially international students, I don’t know what’s going to happen—so I’m afraid a human TA will judge me. So I’m afraid to ask questions.”

Currently, AristAI is in pilot mode with several Illinois instructors expressing interest and testing the tool and hoping to employ it in the upcoming fall semester. The tool is being designed to be embedded within campus educational technologies such as Canvas, MediaSpace, and Piazza. The AristAI website allows anyone to create an account and demo the product. Jiheng and the AristAI team are developing an NSF grant and hope to make the AI teaching assistant commercially available. The team is also exploring how multiple AI agents could work as a virtual teaching assistant team, made possible through new tools like the open-source CrewAI framework.

When asked about the current assumptions and practices regarding GenAI on campus among students and teachers, Jiheng felt that many instructors were unaware of the extensive and diverse use of AI by students. “I don’t think we can stop students from using AI. This is the future. It’s better for us to start guiding students to use AI in the right way. For example, I notice many professors ask students not to use AI for their homework, but it turns out that students use AI anyway. We should tell students how to use AI correctly.”

Jiheng highlighted that the flexibility of AristAI allows it to be customized according to the specific needs and preferences of professors. This adaptability ensures that the AI teaching assistant can align with different teaching methodologies and objectives. For example, the AristAI Teaching Assistant team received feedback from professors that the assistant should not directly answer students’ questions but, rather, guide them to solutions. Jiheng notes that the team is fine-tuning the AI teaching assistant so that “whenever a student asks a question it will not directly give them the answer but guide them to the solution like a real TA.” Jiheng believes students can “not only get their answer but also learn while asking the AI teaching assistant.”

AI Assist: Advising, Teaching, and Career Prep for College of Engineering Students

By Robert Baird, Senior Associate Director at the Center for Innovation in Teaching & Learning  
 

Like much of the academic world, The Grainger College of Engineering at the University of Illinois has been exploring the potential of generative AI for teaching, learning, and research. One of the more intriguing administrative uses of generative AI in Engineering is the exploration of an advising chatbot for students. Professor Jonathan Makela (Electrical and Computer Engineering) who serves as the Associate Dean for Undergraduate Programs in the college, has been coordinating these efforts. Seen as a tool to help the very busy human advisors, once fully realized the always-on, endlessly-patient chatbot will help students 24 hours a day in responding to common questions.

Watch this short video where Jonathan talks about Advising Bots.  
 

Engineering’s advising chatbot is being tested using “uiuc.chat,” a web-based OpenAI tool built by the Department of Electrical and Computer Engineering in partnership with the National Center for Super Computing Applications. By providing the advising chatbot with the college-specific guidelines, documents, policies, and web documents related to advising, the design team ensured that the chatbot would be able to respond to student questions quickly and in a conversational manner. Jonathan: “This, I think, is a really powerful use that we're just experimenting with right now in this sandbox where we can control the information that the AI is using so that we can make sure that as the student is interacting with the chatbot, the information that's given back is the information we want the student to receive.” With concerns regarding AI’s reliability and choice of sources, Jonathan finds “what I like about what’s been set up at the NCSA through this uiuc.chat site is that what comes back from the AI is trained on a specific set of information and actually referenced so you can then go and click and it shows you the website where it got that information.”

Student Privacy Safeguards 

Mindful of student and faculty privacy, the design team built uiuc.chat on a campus platform, separate from commercial and more open AI systems, giving “faculty and other people the ability to experiment with this technology in a controlled and safe environment where you can essentially ask it to incorporate whatever body of knowledge you want.” Jonathan explained that the design and use of the advising chatbot was carefully calibrated to the need to protect student information and privacy: “That's one of the concerns that we have in the student services area, especially when we talk with our advisors and staff that are experimenting with open AI. We need to be protective of student information, even for something as simple as the use case of AI helping draft a response email. You wouldn't want to put in the student's email to you because that contains personal identifiable information that we need to be protective of.”  
 

AI-Powered College Advising 

While the college utilizes many methods to communicate with students, the advising chatbot could offer benefits beyond assuming that students will scan websites and read though FAQs and attend informational sessions. With the advising chatbot, students would simply ask questions directly, with the chatbot assembling the information into a conversational response.  
 

For advising professionals and administrators in the college one longstanding issue has been that some students would not seek out advisors or search through college information. For some, asking a chatbot may be easier and less intimidating than searching for the information or asking someone. Jonathan: “We need to recognize that there are situations in which students are going to be less willing or less likely to come into an advising appointment on their own. So, if we can give them the information that they need to navigate our systems in a way that they're comfortable with and willing to interact with, I think that's a win.” However, he is quick to point out that the AI needs to be able to identify times when students should seek the advice of a college professional, and recommend that course of action. He also points out that the correct information to provide depends on the context of the question (e.g., the major the student is in, the phase of the academic semester the question is asked) and something that currently is not well handled by AI.  
 

A More “Empathetic” Advising Chatbot 

While the advising chatbot has no emotions or feelings, the way it is prompted to respond, and its programmed language style, do convey emotions (or lack of them) to human readers. Jonathan noticed that if the advising chatbot replied to questions in a clinical or just-the-facts manner with a simple bullet list the chatbot felt curt and uncaring. They were able to tune the chatbot so that responses were perceived as more empathetic. Jonathan highlighted that “a lot of times in the advising space, there are feelings and emotions at play. And so it is important to be able to prompt the AI to be a little more empathetic in these responses and provide some information that is now in a different tonality.”  
 

Learning With AI in Engineering 

Jonathan hopes to fine-tune the college’s AI resources based on students’ individual needs: “How [AI] can be used in the classroom changes, whether you're working with first-year students versus Ph.D. students. I think what we're really trying to understand now is how that changes throughout the student experience . . . maybe we shouldn’t be using AI at the lower levels because this is where you're really trying to develop those foundational skills . . . but, then, later on in your career, the learning process becomes more open to using AI to offload some of that lower level work that you now understand, and you can focus in on some of those deeper learning topics.”  
 

Another hope is to teach students to be critical consumers of AI outputs: “I think one of the interesting use cases for teaching, is using AI to ask it a homework problem as part of a classroom activity and then work with your students to critique the results, the answer that AI gave you, because it is not 100% correct . . . if you as a learner can go through and critique and see, yes, this is where it's right, this is where it's wrong and this is why it's wrong, I think that's a really nice demonstration of learning concepts.”

Teaching In a World With AI 

When asked his advice for instructors adapting to the prevalence of AI, Jonathan had some tips: “For a week before classes start, run your homework through ChatGPT and look at the answers that come back. I did that and it was eye opening for me because it provided reasonably correct responses, which was impressive. It knew all the parlance of the domain language and could pull off some of the equations, but it wasn't exactly correct. I would give it a ‘C+’. But if you are a student going through the material for the first time, it sounded 100% plausible. And so that, again, is this idea of having the teacher in the loop to be able to guide students, to understand that this is what's right, this is what's wrong, and this is why it's wrong about that. That can be a valuable part of the learning process.”

Watch this short video where Jonathan talks about how teachers can get started with AI.

Teaching and Learning With AI 

While obviously a new, disruptive technology, AI recalls previous historical moments and traditions: “We’re always going to offload calculations onto computers. So now that the computer incorporates AI, and you can interact with it in a different way . . . You could always go look up the even numbered problems, which were there in the back of the books I learned from, and now you could simply ask the AI what the answer is . . . This is one of the challenges that we have as educators. We have to be able to convince and show the value to our students of the process of learning rather than only focusing on getting the right answer.”  
 

Engineering Careers and AI 

Of course, the disciplinary and career focus of Engineering requires that the college not only pay attention to AI, but that it be actively involved in its use, on campus and off. “I think that it's just a fact that as our students graduate and go off into industry or academic careers, they're going to be utilizing AI with their peers and with their colleagues. So, they're going to need to be conversant in the technology, which is the reason why I think we can’t shy away from its use in the classroom, because it is just another tool that successful engineers are going to utilize in one way or the other. And so, it's important that in the safe confines of the university system our students can experiment with it.”

Did you like this story? Read more about how University of Illinois faculty are using generative AI in the classroom.