HCII PhD Thesis Proposal: Christine Kwon
When
-
Where
NSH 3305 + Remote
Description
Designing Productive Friction in AI Tutors From Last-Mile Learners’ Metered Mindsets
Committee:
Amy Ogan, Co-Chair, Carnegie Mellon University
John Stamper, Co-Chair, Carnegie Mellon University
Jodi Forlizzi, Carnegie Mellon University
Seiji Isotani, University of Pennsylvania
Judith Odili Uchidiuno, Georgia Institute of Technology
Abstract:
While educational technologies (EdTech) hold promise in expanding access to education for learners worldwide, many systems are designed for high-infrastructure contexts and often overlook learners in last-mile environments, which are characterized by limited access to mainstream education, low-infrastructure contexts, and geographic marginality. At the same time, AI holds extensive potential to bridge both educational and digital gaps, providing learning benefits for inclusive and scalable education. While prior research often measured performance, engagement, and adoption metrics to evaluate EdTech solutions in last-mile contexts, how learners actually engage with such systems remains underexplored. Without this grounded understanding, AI risks critical misalignment with last-mile realities and can exacerbate existing divides.
In addressing this gap, my research draws on a multi-year collaboration with an educational technology based in rural Uganda, Yiya AirLearning. They are an offline learning platform that delivers remote courses through radios and basic keypad phones. My prior work investigated learner interactions navigating these technologies from two main studies. In a large-scale qualitative study, learners shared how the radio and phone learning ecosystem provided safe, adaptable, and meaningful learning environments, showing how these technologies align with their contexts. Yet, they also expressed how their preferences, contextual influences, and decisions shaped their use of radios and phones during the course. With this grounded understanding, I then investigated learner interactions with AirVoice, a voice-based mobile AI teacher used to support learning from the main radio- and phone-based courses. From interviews with AirVoice users, learners demonstrated metered mindsets, carefully managing resources like airtime that led to moments of productive friction and intentional interactions with the AI teacher, requiring deliberate decision-making on how and when to interact with the mobile teacher.
Moreover, last-mile contexts are currently framed as spaces for downstream technological integration. However, through the lens of reverse innovation, last-mile contexts can be understood as sources of innovation where naturally emerging friction and practices offer insight into meaningful engagement with AI beyond last-mile contexts. My proposed work aims to conduct targeted investigations with AirVoice on understanding naturally emerging moments of friction and habits in last-mile environments and resulting learners' metered mindsets. I plan to show how they can inform future AIED design for last-mile contexts, and more intentional and deliberate decision-making interacting with AI tools. I will continue my research partnership with Yiya AirLearning to conduct a mixed-methods investigation consisting of three parts:
RQ1: What educational opportunities does interacting with AirVoice reveal beyond implementation value? I will analyze direct conversations between learners and AirVoice to identify the types of learning opportunities and practices they already engage with and determine areas that can be improved with AI literacy support. In conducting this analysis following my prior work, I will provide a more comprehensive understanding of learners' interactions with AirVoice.
RQ2: How would learners carry metered mindsets into their interactions with AirVoice? I aim to investigate how learners potentially carry their metered mindsets into interacting with AirVoice in future courses through various analytical processes. I will collaborate with my partners in Yiya AirLearning to collect and analyze AirVoice data that captures various dimensions of their interactions with the AI teacher, including platform data, chat transcripts, survey data, and targeted pre- and post-course interviews.
RQ3: How would learners’ metered mindsets inform the use of desirable difficulties for learning and interacting with AI? Based on learner-driven insights from my prior work and Phases 1 and 2, I will create a framework that provides clear design implications and opens up research and design spaces for future AIED design for both within and beyond last-mile contexts. I will then apply this framework to AirVoice to validate the framework.
Link to Proposal Document: Christine Kwon Thesis Proposal Draft