BIOCBOT: A GenAI study buddy tool

I had been working as a part of a team to broaden the reach of the highly successful study buddy project. Dr. James Enns in Psychology had been matching students in his advanced statistics course after the first midterm, into study buddy partnerships and had been following their learning trajectories over the rest of the semester. He had been manually matching learners in his class for three years, when one of his students, Achol Jones, had the idea that this could be done in other courses at scale. She brought the study buddy idea to Dr. Warren Williams and me, and we got together a fantastic team including Dr. Patrice Belleville from computer science – which was funded – thank you Faculty of Medicine!

It was through the grant writing process that we were introduced to Dr. Bowen Hui at UBCO. Dr. Hui had built a remarkable tool called “Teamable Analytics”, and we started down the path of adapting her Canvas-compatible tool to our study buddy program. Meanwhile, we were using Qualtrics to match learners in very large enrolment courses in biochemistry and were collecting loads of data on efficacy of student matching. At this point Dr. Enns’ postdoctoral fellow, Dr. Veronica Dudarev, dug deeply into the statistics and meaning of these data. One thing really stood out to Dr. Dudarev, during all these semesters offering the study buddy program — there really was no downside to participants.

No matter where you are on a particular course, you will do better in the course if you work with another student. Our data indicates that students are significantly improving their overall course scores when they participate in the study buddy program, which naturally leads to increased time spent learning-by-teaching. So then, why was only ~30% of our class signing up for the study buddy program? Student exit surveys indicated structural barriers and changing motivation for a particular course as central themes and reasons for learners electing not to participate. We also suspect some resistance of students who are worried that their questions will sound too trivial to one of their classmates. At this point, Daniel DeHeer-Amissah joined the team! Daniel at the time was in Dr. William’s BIOC 202 course, and he was keen to find a way to encourage more students to engage with the study buddy program. This motivated the current initiative. Could we bring the study buddy advantage to students who felt they couldn’t participate in the legacy study buddy system due to situational barriers?

Once BIOCBOT is deployed, I will invite students to participate in one of two tracks:

Out of approximately 600 students, 30 will be selected for BIOCBOT access. We want to evaluate how BIOCBOT serves students across the entire performance spectrum. Students who are struggling may benefit significantly from personalized AI support. However, high-performing students may also benefit immensely, as BIOCBOT could help them teach more effectively and through the teaching of an AI-agent, reinforcing their own learning comparably to the effect we observe in the human peer-study buddy system.

We wrote up a small TLEF grant which was funded – thank you so much TLEF! But more importantly, we were connected to the new LTIC incubator space and had the chance to work with an incredible team led by Rich Tape. We brainstormed how to build an app that would act like a real study buddy. One that would take on the role of learner or tutor, depending on the confidence and current understanding of a particular topic or unit of the learner. It’s really cool!

Key feature: Students can toggle between protégé and tutor mode, and the BIOCBOT will switch between the two modes: asking the learner questions, and summarizing and answering questions. At the moment the instructor is able to add personality to the bot through modification of the prompt.

A teaching opportunity in mind: One nice feature of the way BIOCBOT has been built, is that it is conversational, and it ends each session with an invitation for learners to teach the BIOCBOT something they’ve learned. This “teaching mode” reinforces learning through explanation, even for learners who were working in the protégé mode. So, all learners will spend some time learning-by-teaching when they engage with the BIOCBOT.

The chatbot is easily customizable, with instructors maintaining full control over content. Instructors can upload:

BIOCBOT cites itself based on the instructor’s provided notes. The instructor needs to specify the course duration (number of weeks and classes per week), and BIOCBOT will automatically divide the content into units accordingly.

Note: It is important for instructors to carefully plan their course structure and ensure learning outcomes are clear and specific for optimal use of BIOCBOT.

We are really looking forward to getting this tool in front of live learners in BIOC 202 this semester, and are very excited to hear all about their experience and honestly, to see how the BIOCBOT compares in terms of learning trajectory relative to the students who participate in the legacy study buddy project. Is being able to study from your own home with an AI buddy, and on your own schedule, going to outweigh the learning that occurs when two motivated scholars discuss the content and come up with questions and concept maps together? Only time will tell.

We are optimistic that students will benefit from BIOCBOT and its built-in “teaching mode” prompts at the end of each session. When you are teaching a bot, do you have the same feeling of responsibility for the clarity and accuracy of the content? Is this one of the reasons why the legacy program is so effective? Will this efficacy be reduced when students are working with a GenAI study buddy? Stay tuned for more!