Learning to Learn With AI — The Texas Scientist

By Marc Airhart. Illustrations by Kouzou Sakai.

Since the advent of smooth-talking, smart-sounding AI chatbots like ChatGPT, educators have fretted about how generative AI might impact teaching and learning.

But what if AI could act as a force multiplier for important work already underway in the classrooms of the most dedicated instructors? What if it could satisfy some of the demand for more customized, one-on-one interactions? Or if it could meaningfully help students who arrive in class with gaps in their prior knowledge or who struggle more than others to learn new material? Rather than fear AI or ban it from classrooms, University of Texas at Austin faculty members are finding ways to augment and improve course instruction with new tools.

The College of Natural Sciences, through its Office of STEM Education Excellence (STEMx), has given grants to about a dozen faculty members to explore how best to use a host of AI applications in the classroom. One project helped an instructor identify where her students were struggling with concepts, while another helped students explore new ways to search and understand scientific literature. Meanwhile, the University is building a new AI tutor called UT Sage that can be trained on the content and goals of any specific course and patiently help any student at any time.

Clear as Mud

Students learn best when they experience “metacognition,” or thinking about their learning process and identifying what they don’t understand. To help students get into this mindset, teachers often use a strategy called “muddiest point,” which involves surveys to find out what concepts tend to trip up learners.

Ann Thijs, a faculty member in biology, teaches an upper-level ecology course and polls her undergraduates each week to find out where they struggle. Generative AI tools help her analyze the results, uncovering trends about the concepts students find most challenging, and then also help her develop a related “muddiest point” handout draft. All of this has helped supplement regular lectures and textbook learning in ways that fill in students’ knowledge gaps.

In an end-of-class survey, many students told Thijs they “appreciated it, felt heard and felt like it helped.” This AI-assisted approach could be used in virtually any kind of class, layered on top of traditional classroom teaching or to help teaching assistants set the agenda for collaborative study sessions with students.

“It’s a widely applicable implementation that’s easy to use and doesn’t require a lot of technical skills,” she said.

It’s All Connected

If you want to be a scientist, you need to know how to do a literature search. After all, to design a useful experiment or build a new theory requires knowing what’s already known, what’s still unknown and how others have tried filling in the gaps.

To teach this skill, K.P. Procko assigns the students in her introductory biochemistry course a culminating project in which they explore the research literature and create a scientific poster about a specific disease. Students sometimes struggle to find the most impactful papers. They don’t know how the field has progressed historically, who the big players are/were and which advances were most influential.

Last fall, Procko, an associate professor of instruction, introduced students to an AI tool that creates a visual network map of related research papers.

“Instead of being faced with a wall of manuscripts after a keyword search, students have this visual network that they can click on and explore other papers,” Procko said.

Nearly two-thirds of the students reported that they preferred the results of the AI tool to a conventional keyword search because it was less overwhelming. But they acknowledged there was give and take, with some leveraging of both strategies yielding the best results. Procko said the exercise is about more than this one tool; it’s getting students to critically evaluate the strengths and weaknesses of new technologies and learn to use them alongside more traditional methods.

“Our main takeaway was that by using AI in a structured way, students get skeptical of its use, which is exactly what we need,” she said. “We need to use our brains as we work with these tools, because they’re still very valuable.”

Your Own Personal Tutor

A lot of things can get in the way of learning in college. Maybe you’re struggling to learn a concept in organic chemistry on Sunday afternoon, but your professor isn’t available until Tuesday. Or your high school chemistry class didn’t prepare you for college-level work, and now on day one, you’re already behind. Or you’re afraid to raise your hand in class and admit you’re lost.

UT’s Offices of Academic Technology (OAT) and Enterprise Technology have developed a responsible GPT-based AI tutor called UT Sage that is trained on the material for each specific course in which it’s used by the faculty member who is teaching that course. Julie Schell, assistant vice provost of academic technology and director of the OAT, conceived of the original vision of the tutor, which is available 24/7 and is infinitely patient and supportive. It’s like an inexhaustible virtual teaching assistant, ready to go as many rounds of questions as a student needs to ask.

Unlike a typical chatbot focused on answers, UT Sage is designed to respond to prompts from students in a Socratic way, helping students think through problems and making sure they understand, the way a good professor does. Like Ann Thijs’s muddiest point experiment, it’s designed to boost metacognition, helping students evaluate their own learning process.

“The Holy Grail is self-directed learning,” Schell said. “We’ve baked this concept into Sage. Sage doesn’t start by telling you all of the answers, but what it does is coach you through thinking about questions that you might want to think about if you’re going to learn and perform certain tasks.”

Sage is currently being tested by select faculty members in classrooms across the University, including in the College of Natural Sciences, with the goal of eventually making it available for any and all classes.

“Using responsible AI-based technology to solve teaching and learning problems that we cannot otherwise solve,” Schell said, “is what excites me most about Sage and generative AI more broadly.”