Rural Indian Students Share the Same AI Barriers, Study Finds

Artificial intelligence is sweeping through university lecture halls across the world, but a new study from India suggests that in rural higher education, the obstacles students face are strikingly uniform. Research published in SN Social Sciences by Dinesh Satapathy, V. Sri Divya, Deepak Kumar Pradhan and Arpan Kumar Nayak examined how students in the eastern state of Odisha perceive the barriers standing between them and meaningful use of AI tools. The central finding is counterintuitive: it is not who the students are that shapes their difficulties with AI, but where and under what conditions they study. Gender, age, field of study and family type barely moved the needle once the researchers applied rigorous statistical corrections, pointing instead to shared contextual conditions as the dominant force behind AI hesitancy in rural campuses.

The study arrives at a moment when AI adoption in higher education has become one of the most intensively researched topics in educational technology. Bibliometric analyses show an explosion of publications on generative AI in universities, and systematic reviews have catalogued factors influencing student readiness, from technological self-efficacy to prior digital literacy. Yet the authors note a conspicuous gap: most of this evidence comes from urban, well-resourced institutions in North America, Europe and East Asia, or from major Indian metropolitan centers. Rural contexts, where infrastructure is thinner and digital divides run deeper, remain empirically underexplored. India’s own census classifications define rural areas in ways that shape access to electricity, broadband and institutional funding, making the rural-urban distinction far more than a demographic label. It is a structural boundary condition for technology adoption.

To probe this boundary, the team employed a quantitative survey design targeting rural higher education students in Odisha. Data were collected from 111 students using a ten-item instrument built on a Likert scale, with each item probing a different facet of perceived barriers. The researchers organized these facets into five conceptual clusters: access constraints, meaning unreliable connectivity and inadequate devices; knowledge gaps, meaning insufficient training in how AI systems work; attitudinal resistance, meaning skepticism or discomfort toward AI; cognitive concerns, meaning worries about overreliance and the erosion of independent thinking; and ethical apprehensions, meaning anxieties about privacy, bias and academic integrity. This multidimensional structure matters because treating AI barriers as a single blob obscures the different levers policymakers might pull to address them.

The psychometric quality of the instrument was strong. The scale achieved a Cronbach’s alpha of 0.912, a measure of internal consistency that indicates the ten items reliably tapped a coherent underlying construct of perceived barriers. Values above 0.9 are generally considered excellent in psychometric theory, lending credibility to the composite scores the researchers analyzed. Rather than relying solely on classical parametric statistics, which can be fragile with modest sample sizes and ordinal Likert data, the team used non-parametric tests for group comparisons and reported effect sizes alongside p-values. This dual reporting is important because statistical significance in a small sample can be misleading without a sense of how large the differences actually are.

One of the most methodologically notable features of the study is its handling of the multiple comparisons problem. When researchers test many items across many demographic groups simultaneously, the probability of finding at least one spurious statistically significant result inflates rapidly. To guard against this, the authors applied the Benjamini-Hochberg procedure, a widely used technique that controls the false discovery rate, the expected proportion of false positives among all declared significant findings. This correction is often skipped in educational survey research, which means many published group differences may not survive scrutiny. By applying it here, the Odisha team set a higher evidentiary bar, and the results they report are more trustworthy as a consequence.

What did the corrected analyses reveal? In short, demographic segmentation largely dissolved. Gender, age, field of study and family type did not demonstrate statistically robust differences in perceived barriers after the false discovery rate correction, and the effect sizes were generally small, indicating negligible practical differences between groups even where nominal differences appeared. The one demographic variable that showed any differentiation was academic level, and even there the signal was limited and marginal. It surfaced as a single item-level difference and a small but statistically significant difference in the composite summary index, suggesting that students at different stages of their programs may experience slightly different intensities of barrier, perhaps reflecting growing familiarity with academic technology demands as students progress.

The interpretive weight of the study rests on what this null-heavy pattern means. The authors conclude that perceived barriers to AI integration in rural higher education are shaped more by shared contextual conditions than by demographic characteristics. In other words, a female undergraduate in the humanities and a male postgraduate in the sciences at the same rural institution are likely to report similar obstacles: the same patchy internet, the same lack of formal AI training, the same institutional culture of limited exposure. This finding aligns with a broader literature on digital divides, which has increasingly emphasized that structural and situational factors, rather than individual attributes alone, determine how people experience and adopt new technologies. It also echoes work on low socioeconomic digitalization, where citizens’ difficulties with digital services were traced to situated conditions rather than personal deficits.

The theoretical scaffolding of the study draws on the technology acceptance tradition. Fred Davis’s Technology Acceptance Model, introduced in the 1980s, proposed that perceived usefulness and perceived ease of use govern whether people embrace new information systems, and the later Unified Theory of Acceptance and Use of Technology extended this framework with additional moderators. Studies applying these models to AI chatbots and educational AI have repeatedly found that prior AI experience and digital literacy moderate acceptance. The Odisha findings add a rural inflection to this tradition: when the environment supplies little prior exposure and little infrastructure, the usual demographic moderators that animate urban studies may be swamped by the sheer weight of shared disadvantage. The barrier profile becomes a property of the setting, not of the population within it.

For policymakers and university administrators, the practical implication is that institution-wide strategies, not targeted demographic interventions, are the appropriate response in rural settings. If barriers are common across gender, age and discipline, then programs aimed narrowly at, say, women students or science majors would miss the bulk of the problem. Instead, the authors underscore the importance of strategies that operate at the level of the whole institution: improving connectivity and device access, embedding AI literacy across curricula, addressing attitudinal and ethical concerns through transparent institutional policies, and normalizing critical engagement with generative AI in classrooms. Related research on pedagogical approaches to generative AI has argued that fostering critical engagement, rather than mere technical skill, is key to preparing students, and the rural context makes this doubly urgent because students there often lack informal exposure that urban peers acquire organically.

The study’s authors are candid about its preliminary nature. A sample of 111 students from one Indian state cannot settle questions about AI adoption everywhere, and self-reported perceptions are not the same as observed usage behavior. The composite summary index difference by academic level hints at dynamics that larger, longitudinal designs could unpack, tracking whether barriers attenuate as AI tools proliferate or harden into durable inequalities. The dataset and instrument are available from the corresponding author upon reasonable request, enabling replication. Still, the core contribution stands: rare context-specific empirical evidence from rural higher education, gathered with unusually careful statistics, showing that the AI divide in places like Odisha is not a mosaic of demographic micro-divides but a single shared landscape of constraint. Bridging it will require acting on the landscape itself, not on the demographic variations that cross it.

Subject of Research: Perceived barriers to artificial intelligence adoption among rural higher education students in Odisha, India

Article Title: Perceived barriers to adoption of artificial intelligence among rural higher education students

Article References: Satapathy, D., Divya, V. S., Pradhan, D. K., & Nayak, A. K. (2026). Perceived barriers to adoption of artificial intelligence among rural higher education students. SN Social Sciences, 6(10), Article 500. https://doi.org/10.1007/s43545-026-01801-1

Image Credits: AI Generated

DOI: 10.1007/s43545-026-01801-1

Keywords: artificial intelligence, higher education, rural students, digital divide, Odisha, AI adoption, technology acceptance, survey research, perceived barriers, educational technology, India, Benjamini-Hochberg