Hyejin Park, a master's student in the Learning Analytics and Artificial Intelligence (LAAI) program, was invited to present her research at the Practitioners, Industry, and Policy (PIP) track of the International Conference on Artificial Intelligence in Education (AIED) 2026. In a particularly competitive year, only nine papers were accepted from 40 submissions, representing a 22.5% acceptance rate.
For Park, whose paper examines South Korea's AI Digital Textbooks (AIDTs), the invitation marks an important milestone as an emerging scholar and practitioner.
“This is my first international conference, which makes it extra meaningful,” she said. “I chose the PIP track deliberately, because I wanted to bring a practitioner's voice into the conversation.”
Her research explores a critical question in AI-powered education: how much agency do students and teachers actually have within systems marketed as personalized learning tools? The paper introduces a framework called Decision-Intervenability to analyze how AIDT interfaces distribute control among students, teachers, and the system itself.
Applying the framework, Park and her collaborators identified seven distinct types of control within the platforms. Their findings revealed that “a domain-driven pattern that consistently prioritizes system and teacher authority over student agency” exists across all publishers' products. In fact, three-quarters of learning modules provide students only what the researchers describe as “shared control,” where learners operate within predetermined pathways rather than directing their own learning.
“What struck us most is that the same pattern holds across all publishers,” Park said. “That suggests it isn't a matter of individual vendor architecture choices, but a reflection of the government guidelines they all built under.”
The project grew out of Park’s own experience as an educator.
“As a teacher, I noticed that the ‘personalized’ tools coming into my classroom adjusted the pace and sequence of content for students, but rarely let students adjust anything themselves,” she said. “The system made every decision, and the learner simply followed.”
While current AIDTs can effectively deliver content and assessments at scale, the study raises broader questions about whether personalization alone supports meaningful learning.
“The promise is personalization, but the underlying structure is one of constrained agency,” Park said.
She also said that it was particularly meaningful that this year’s AIED conference was held in South Korea, the very context her research investigates.
“When the paper was accepted, I realized this could resonate with a much wider audience beyond just us,” she said. “And because this conference is being held in Korea, I was even more thrilled that we'd get to share this work with people who are close to the very context we're studying.”
Park credits Penn GSE faculty with helping shape both her research and perspective. The project began in Haiying Li's course, “Large Language Model Applications in Education,” and evolved into a conference paper through Li’s continued mentorship.
“Her guidance sharpened how I reason and write: how to weigh a question from the perspectives of the different stakeholders it affects, and how to build an argument that is both rigorous and persuasive,” said Park. “This work wouldn't have reached publication without it.”
Looking ahead, she sees the conference as more than a presentation opportunity.
“I'm no longer just using edtech tools, but questioning them, and doing it in a room where those questions can actually shape what comes next,” she said.
Ultimately, her work centers on a challenge she believes will only grow more important as AI becomes increasingly embedded in education: “How much agency and ownership does the learner hold, and how do we balance that with the technology? That balance isn't something that happens; it's something we design, with real stakes for the learner.”
Media Inquiries
Penn GSE Communications is here to help reporters connect with the education experts they need.
