Turning to the text box: How LLMs are used by first-years taking 8.01
Students and instructors in Physics I (8.01) discuss how AI has impacted how the class is taught
Learning classical mechanics has long been a source of frustration for first-years at MIT. Even for those with high school experience in physics, understanding certain concepts was nearly impossible without dedicating extra hours to the subject. For some students, the rise in artificial intelligence (AI) is making studying more efficient. Large language models (LLMs) give students unprecedented ease of understanding arcane concepts in a fraction of the time it would take them to do alone.
The Tech interviewed current students and instructors in Physics I (8.01) to understand the influence this technology has had on how the class is taught.
To “further understanding” and “deepen knowledge”
The homework assigned to students in 8.01 consists of learning sequences and problem sets. As one student explained, their four weekly learning sequences serve to give them a preview of content they will later learn in class. (This student will be referred to as Student A, as they requested to remain anonymous. The other interviewed students will be given pseudonyms for similar reasons.)
“Overall, I think they have helped my understanding of the content,” Student A said, “but I also think that they can be a bit time-consuming to do, especially for an MIT student who just generally has a bunch of other demands and extracurriculars to attend.”
Despite the burdensome amount of time it takes to complete learning sequences, Student A stated that they never used AI on these assignments; instead, they tended to use AI as a guide for their problem sets when necessary. If they spent half an hour stuck on a question, for example, they would ask Google Gemini what topics would be necessary to begin solving it. If the student was still stuck, they would go back to Gemini with their attempt at the problem. “Here is my current attempt at solving it,” Student A reported typing. “Do not give me the answer; give me a small hint about how to improve my approach.”
This sentiment was echoed by other interviewed students. Student B, when asked about their AI usage in this class, stated that they use ChatGPT to complete their learning sequences in a similar manner.
For Student B, the Technology-Enabled Active Learning (TEAL) learning style in 8.01 is too fast-paced. The two-minute videos in the learning sequence are not enough for them to understand the questions that followed. “Sometimes, I like to use AI to further my understanding and deepen my knowledge in the particular topic that we’re being taught,” they said.
These students reported knowing that simply asking the chatbots for the solutions would be short-sighted. They recognized that problem set questions were considerably similar to those on their exams, meaning it would be in their best interest to complete their problem sets as independently as they could.
However, the availability of AI has affected their reliance on peers and office hours in the class. For instance, Student C admitted that they ask Gemini for help with problem sets about as often as they ask their peers. “I like to do my [problem sets] pretty early on in the week,” they explained. “A lot of my peers don’t start the pset on the weekends.”
Other interviewed students cited ease of access as a reason they turn to AI as an alternative to office hours. As Student A explained, however, while prompting Gemini for help is much more ideal than traversing campus to attend office hours, that was not the only reason AI is often more convenient.
Since 8.01 has hundreds of students, “there can be a lot of other people at office hours,” Student A said. “Sometimes, when I’m really stuck on a problem, it can feel like I’m kind of hogging the [teaching assistant] or instructor’s time. But with an AI, I can just type in as many dumb questions as I want, and I never feel like I’m wasting anyone’s time because it’s a machine.”
How 8.01 is adapting to AI
The instructors of 8.01 are well aware of the influence AI has on their students. To ensure that students are still learning the fundamental concepts of classical mechanics, the instructors tasked themselves with adapting their policies to best serve everyone.
Since last year, the 8.01 Canvas page has featured an AI policy, which emphasizes that chatbots should not be completing work for the students. “We expect that you will find working with your peers more useful than working with AI,” the policy reads, “but if you want to use AI as a part of getting unstuck after first working on your own, that is fine.” The policy also notes that students must list any AI tool they used as a collaborator on their problem sets to maintain transparency.
Instructors Michelle Tomasik and Krishna Rajagopal offered further comment on how AI has (and has not) affected teaching 8.01. Despite the availability of LLMs, Tomasik and Rajagopal stated that 8.01 still offers 23 hours per week of office hours for students to attend. They reported that their office hours tend to be “full of students,” just as interviewed students described them. The instructors appeared to have the same general expectation of AI usage as their students, as they emphasized how students should first attempt problem sets independently before asking a peer or AI.
“If a student relies upon either peers or AI [too] much, without first working through each problem themself, they will not build strong problem solving skills and will not do as well on the exams, which contribute five times as much to their grade as do the Problem Sets,” Tomasik and Rajagopal explained.
More than just classical mechanics
The testimonies from 8.01 students and instructors reveal something important about learning in the age of AI: if used responsibly, AI can contribute to a net positive. With clear communication of policies and expectations, students can readily access assistance when they need it, especially in larger classes.
AI is not going anywhere anytime soon, which leads many to believe that the optimal option is to make the best use of it. As Tomasik and Rajagopal put it, “it is important for students to learn how to engage effectively and productively with AI tools during their four years at MIT.”