Mapping the AI Divide in Undergraduate Education: Community Detection in Disciplinary Networks and Survey Evidence

TL;DR

The paper addresses inequities in undergraduate AIGC literacy by mapping disciplinary structures at Nanjing University using a network-of-courses approach and survey data from 301 students. It identifies four curriculum-based communities and shows that motivational efficacy—especially skill efficacy—partially mediates differences in AIGC literacy across communities, while usage efficacy has a weaker mediating role for evaluation. The study integrates a greedy modularity community detection framework with MAILS-inspired literacy measures to reveal how curriculum similarity shapes motivation and literacy, offering a scalable institutional design for diagnosing and reducing the AI divide. The findings highlight the importance of cross-disciplinary integration and longitudinal, curriculum-wide interventions to promote equitable AI fluency and responsible use in higher education.

Abstract

As artificial intelligence-generated content (AIGC) reshapes knowledge acquisition, higher education faces growing inequities that demand systematic mapping and intervention. We map the AI divide in undergraduate education by combining network science with survey evidence from 301 students at Nanjing University, one of China's leading institutions in AI education. Drawing on course enrolment patterns to construct a disciplinary network, we identify four distinct student communities: science dominant, science peripheral, social sciences & science, and humanities and social sciences. Survey results reveal significant disparities in AIGC literacy and motivational efficacy, with science dominant students outperforming humanities and social sciences peers. Ordinary least squares (OLS) regression shows that motivational efficacy--particularly skill efficacy--partially mediates this gap, whereas usage efficacy does not mediate at the evaluation level, indicating a dissociation between perceived utility and critical engagement. Our findings demonstrate that curriculum structure and cross-disciplinary integration are key determinants of technological fluency. This work provides a scalable framework for diagnosing and addressing the AI divide through institutional design.

Figures (3)

• Figure 1: Hypothetical model
• Figure 2: Nodes represent various schools of Nanjing University, and the edges denote curriculum similarity. Four communities are divided based on the characteristics of each school, with the representative school of each community being the one with the highest degree centrality within that community. Among them, Community 1 is a science-dominant community, represented by the Kuang Yaming Honors School; Community 2 is a science-peripheral community, represented by the School of Life Sciences; Community 3 is a science & social sciences community, represented by the School of Information Management; Community 4 is a humanities and social sciences community, represented by the School of International Studies.. The degree centrality of the four communities decreases sequentially from 1 to 4.
• Figure 3: Four communities are divided based on different schools at Nanjing University. Community 1 is a science-dominant community, Community 2 is a science-peripheral community, Community 3 is a science & social sciences community, and Community 4 is a humanities and social sciences community. The degree centrality of the communities decreases sequentially from 1 to 4.