Don’t fret, not all AI software will “upend” the classroom. Chatbots overseen by faculty can improve grades for more engaged students as they settle into college.
Georgia State University undergraduates with access to a course-embedded virtual assistant were more likely to earn a passing grade and seek supplemental instruction, according to a new study.
Researchers from the Brookings Institute, Brown University and five other universities examined student behavior across two large, asynchronous courses required for graduation. The study targeted an online environment where students had fewer chances to connect with peers and instructors.
Nearly 2,500 students (63% freshmen) received proactive AI outreach. They were:
- Four percentage points more likely to earn an A or B in a course
- More likely to complete assignments
- Less likely to drop a course or receive a non-passing grade.
Pell-eligible and first-generation students were just as likely to succeed at higher rates.
The chatbot texted students two to three times a week with due-date reminders, personalized feedback on class progress, referrals to academic resources and a hotline to a course instructor. Students who fell behind were nudged to complete assignments.
Your next read: How this university reduced data siloes to maximize AI’s potential
Proactive outreach helps young residential students adapt to living independently while balancing coursework. Targeted help also helps learners navigate the “hidden curriculum” of course requirements and office hours, said Katharine Meyer, a fellow in the Brown Center on Education Policy at Brookings Institute.
Most students were engaged with their virtual assistant. More than half (60%) found the messages helpful, and 82% would recommend a chatbot for this course and others. Less than 5% opted out of messaging.
Keeping faculty in the driver’s seat
Georgia State gave faculty control over content delivery. Virtual assistants functioned primarily to help with administrative tasks and steer students toward learning support.
As a result, learners were also more likely to attend university tutoring. This is a crucial finding as previous assessments on student AI use have found that it can decrease participation in office hours and tutoring.
Learn more about implementing your own AI agent in our next webinar
Unlike generative AI models that create (and hallucinate) new material, Georgia State’s chatbot was trained exclusively on course data. Any question a student asked outside that scope was flagged for a teaching assistant to review.
Limiting what data is available to a chatbot becomes more crucial as AI continues to evolve. One course used the chatbot to generate multiple-choice practice quizzes with questions created by a teaching assistant.
“What the tools can do and what they should do are two different questions; you’re always trying to balance those two things,” says Lindsey Fifield, director of Student Success Engagement at Georgia State University. You want to ensure that “student data is protected and that the autonomy over how the course is taught still rests with the faculty.”
