Using AI to Facilitate Self-Regulated Learning (Kevin Gregorio)

Kevin Gregorio | LinkedIn

Guest Writer

Jun 24, 2024

Image generated by Dalle 3 of a student shaking hands with an ai and surrounded by icons symbolizing different school subjects

Kevin Gregorio is the Director of Learning Services at St. Joseph’s Preparatory School in Philadelphia. Currently pursuing a PhD in Educational Psychology from The Philadelphia College of Osteopathic Medicine, Gregorio is passionate about and actively researching retrieval practice as a means of more effective learning. Ever since the Prep partnered with Flint in February 2024, Kevin has seen the ability of teachers to use Flint to encourage positive student outcomes and for students to use Flint to drive their own learning and leverage the perks of retrieval practice.

Retrieval fosters retention

Studies have shown that when you test students they retain more information, for longer, and in more meaningful ways than if you don’t test students. This is especially true if you test students consistently. Researchers have known this since the 1970s and it’s called the Testing Effect. Since the formulation of the Testing Effect, subsequent researchers (Karpicke, 2017; McDermott, 2021; Yang et al, 2021) have found that it doesn't necessarily need to be a test in order to produce the same effect. In other words, it doesn’t have to look or feel like a test to still get the benefit of a test. For example, instead of a traditional sit-down test, a student could try to explain the content of an upcoming exam to his parents, or to a teacher and benefit from the so-called Testing Effect. It only requires that students attempt to recall and explain the information from memory without assistance from notes or anything that might help them recall what they are trying to explain. Researchers call this Retrieval Practice.

Adapted from Dalle 3: “Student studying using flashcards, with a happy but concentrated look on their face.”

Despite an enormous sum of articles demonstrating the effectiveness of Retrieval (and the Testing Effect) few teachers know about it and even fewer students choose to use it on their own. In fact, current research on retrieval is very much interested in this exact question: how do we get students to use retrieval on their own? Sure, we can implement it into the classroom in a number of ways; but students would benefit greatly from implementing this into their independent study and practice. One form this could take is instead of passively reviewing notes or pages from a textbook, students should quiz themselves on the same material. Unfortunately though, they don’t seem to elect to do that without considerable intervention.

AI for student-driven practice

Recently, however, my work in the classroom with Artificial Intelligence (AI) has given me hope. When first introduced to me in an Ed Psych graduate program, my professor encouraged us to experiment with AI as a self-testing tool. Chat GPT 3.5 proved invaluable in its ability to generate novel problems for me to test myself with (in accordance with the Retrieval literature) along with ample feedback and corrections when I struggled (also in accordance with the literature!).

Created with Dalle 3: “Point of view looking over a student's shoulder to see practice problems on an iPad about the parts of a cell.”

Since then, a number of AI tools have made their way into schools and classrooms across the country. I’ve attended a number of conferences and talks where teachers outline how they’ve used Chat GPT and Large Language Models (LLMs) like it in their classroom; Khan Academy has implemented an AI tutor called Khanmigo; other tools like MagicSchool’s AI has a whole host of amazing tools and resources (especially for teachers!); Quizlet has incorporated two really impressive AI features, namely Q-chat and magic notes; and, most recently in my own experience, a tool called Flint.

While I’ve used and enjoyed all of these tools to some capacity, Flint in particular has stood out. A small start-up based in San Francisco, Flint is marketed as a K-12 AI tool. But having worked with it now for several months, I fear that description belies how profound of an impact I think the tool (and others like it) can make on education as a whole. In large part, because of its ability to facilitate self-regulated learning and Retrieval.

Using Flint to teach retrieval

When I and the teachers I’ve worked with first encountered Flint, we tended to see one obvious application: it’s great for making creative chat-based assignments. It generates the assignment, proposes some rules the AI will follow, proposes entire criteria for grading and feedback, and then, on the back end, gives students and teachers a profile of not only the students’ strengths and weaknesses on the assignment— but it provides a profile of strengths and weaknesses for the whole class. It’s truly profound. Not only does it deliver a highly rich and meaningful activity for the students, it provides actionable insight to the teacher on how to help students develop their strengths and weaknesses. That’s the most obvious application of the tool. But another, perhaps less obvious, application began to dawn on me.

Screenshot of tutor analytics generated by Flint outlining strengths and areas of improvement for the whole class, complete with citations for each performance insight.

Included in Flint is a “Friendly AI” chatbot designed for students to interact with. At first, I was skeptical of this feature for fear that students would simply use it to go directly to answers— to cheat, effectively. And they definitely did that. Once Flint made it possible for teachers to oversee how students were using the Friendly AI feature, that became apparent. But something else became apparent too. Something far less predictable to me. Students started to use Flint to test themselves. Some started to do it on their own. But many more followed them once I started to encourage it in class. They were finally using Retrieval!

Screenshot of an example student session in Flint with the Friendly AI.

Impact of using Flint’s AI for retrieval

What’s more, Retrieval with generative AI isn’t simply a practice test. It’s far more dynamic—and, I think— effective. An AI chatbot like Flint is able to ask students questions, change the difficulty level upon request, provide feedback and scaffolding, and challenge students to go beyond the rote memorization of facts by forcing them to explain concepts in their own words. Furthermore, it’s available to them 24/7, unlike an actual peer tutor or a teacher. If motivated, and taught to use it effectively, students can use these tools as often as they’d like. In other words, they can take control of their learning.

One student, in particular, became so engaged by Retrieval on Flint that he accredits it with getting him off of academic ineligibility. And his usage on Flint speaks to this. At first, he was only using Flint when specific assignments were given to him, but more and more you can see that he started to use Flint during his free time to test himself on classes he struggled in. In fact, he went from having multiple failing grades to having no grades below a 75% in just a few weeks.

Created with Dalle 3: “Student with lacrosse stick being supported academically and intellectually by an AI robot.”

This student’s story, and others like it, is an amazing proof of concept: AI can facilitate meaningful Retrieval and make students self-regulated learners. As a result, Flint, and AI tools like it, give me great hope for the future direction of technology’s role in education.