AI-Powered Teaching Assistants Can Drive Student Success

AI Assistants Can Offer Nuanced and General Guidance

DeOrio and four research assistants used U-M Maizey — the university’s AI toolkit based on custom data sets — to build the chatbot and lecture slides, weaving together threads from past class forums and other course materials to train it.

Using the bot during the distributed systems project was optional. Of the students who used it, 77% said their interactions with it had been helpful, particularly when they posed conceptual and project specification-related questions. The inquiries DeOrio fielded tended to involve more complex topics, he says.

“I would love it if the bot could answer the easy questions, and then they could come to me with the hard ones,” he says. “To me, that says I’m helping students get further along in their educational journey.”

Alexa Alice Joubin, an English professor and co-director of the Digital Humanities Institute at The George Washington University, wanted students in her humanities and critical theory courses to use artificial intelligence for more than just summarizing assigned readings or writing essays.

Piecing together five grants, Joubin worked with a computer science graduate student last summer to create an AI teaching assistant prototype that she customized for both classes.

The open-source, open-access tool can be used to detect patterns in students’ writing. For example, maybe they overuse certain words or phrases, or make repetitive statements.

When responding to a prompt, the AI-powered teaching assistant can also detect whether multiple sources would be applicable to the inquiry to create a comprehensive response.

“It’ll say, ‘I’m breaking it down; I’m going to retrieve answers from this specific document, and because the content seems to connect to these other documents, I’m going to look over there and give you this relevant quote or this passage that is a must-read, so you can take a closer look,’” Joubin says.

RELATED: Here are a few practical ways to adopt Gemini in education.

AI Teaching Assistants Also Can Be Visual

Several years ago, Muhsinah Morris, Metaverse program director and professor at Morehouse College, began using an immersive virtual reality environment to teach an organic chemistry course. The structure allowed her to assist students in real-time as prerecorded lectures played.

Not long after, she learned about functionality that would let students interact with an avatar that could provide a scripted response to questions. Morris worked with the college’s education technology partner to create a 3D, AI version of herself with a computer-generated voice that she says sounds surprisingly like her own.

“It just makes it more familiar,” Morris says. “Especially when you’re working with emerging technologies for students, you want to be able to meet them where they are, and you want them to have a certain level of comfort.”

Morehouse students can engage with the teaching assistant via a Google Chrome browser by typing or verbally asking a question; a captioned response is provided if their laptop’s sound is turned down during class.

The tool can be helpful in settings such as chemistry labs, Morris says, where following the proper procedures and techniques is important. An instructor could use an AI teaching assistant to show students building a distillation apparatus where all of the parts on it should go, for example.

“It gives another layer of autonomy to the students to be able to work more independently,” she says. “Because in those lab settings, you’re talking about 25 students and just one or two people circulating around to make sure they’re doing the right thing.”

Schools Are Employing Data and Other Best Practices

Cybersecurity must be a consideration when crafting a generative AI-powered teaching assistant. For instance, DeOrio didn’t enter private student data into his assistant, and access to the tool was limited to students in his course.

Similarly, predefined boundaries ensure Joubin’s AI teaching assistant draws information solely from files she uploads.

“If we have an assigned piece of reading, but it is not openly licensed, I would use my own words as if I’m teaching it,” she says. “We’re able to have guardrails. If you try to ask things that are not relevant, it’ll say, ‘I’m here to help you with this course; I can only have a conversation about things that are in my data set.’”

AI teaching assistant technology may not be ideal for some tasks. Survey results showed DeOrio’s students don’t feel it’s particularly helpful when testing and debugging their projects; they often come to office hours with questions about that, he says.

“This is definitely not a teaching assistant replacement,” he says. “This has allowed students to make more effective use of the time they have with their professors and teaching assistants.”

SUBSCRIBE: Sign up for the EdTech newsletter for weekly updates.

Hallucinations, which occur when AI provides incorrect information, can also be a problem.

Educators, according to Joubin, should encourage students to question whether the responses AI-powered teaching assistants provide are accurate.

The process, she says, could also help students build the skills they’ll need to identify the most trustworthy, relevant sources of information in the future as the amount of available data continues to grow.

“It could guide us to think more critically,” Joubin says. “Practice makes perfect. AI could play a role because it’s free, it’s available 24-7. Professors are not.”