Generative AI adoption in higher education viewed from around the world

Generative artificial intelligence has swept through university campuses worldwide at a speed that has left institutional policies, faculty training programs, and national education ministries struggling to keep pace. A new analysis published in the journal Higher Education by Yuxi Wen of the College of Education at Michigan State University takes stock of this uneven global transformation, arguing that the integration of generative AI into higher education cannot be understood through the lens of any single country or institutional type. The review of global perspectives, published on 12 August 2026, synthesizes evidence from North America, Europe, Africa, Latin America, and East Asia to show how political, economic, and cultural contexts shape whether and how universities adopt tools such as ChatGPT, DeepSeek, and related large language models. The timing is significant: a UNESCO survey conducted in 2023 found that fewer than 10 percent of schools and universities had formal guidance on AI, but by 2025 a follow-up survey reported that two-thirds of higher education institutions either had or were developing such guidance, a remarkable institutional pivot within roughly two years.

At the technical level, the article situates generative AI systems as a fundamentally different kind of educational technology than those that preceded it. Earlier waves of edtech, from learning management systems to adaptive tutoring platforms, largely automated distribution and assessment. Large language models, by contrast, generate novel text, code, and reasoning in response to open-ended prompts, functioning as conversational partners rather than content repositories. Scholars such as Miao have examined the technical principles underlying these systems, including transformer architectures trained on massive corpora that predict likely token sequences, and have identified both their educational applicability and their fundamental controversies, including hallucination, bias embedded in training data, and opacity about how outputs are produced. These properties cut against the traditional assessment model of higher education, which assumes that written artifacts produced by students reflect the student’s own cognition. When a model can produce a competent essay in seconds, the evidentiary chain between the artifact and the learner’s mind is broken, forcing a redesign of assessment at a scale few institutions have yet attempted.

The empirical literature on student outcomes tells a more nuanced story than either the hype or the panic suggested. A systematic review and meta-analysis by Deng and colleagues published in Computers and Education examined experimental studies of ChatGPT’s effect on learning and reported measurable benefits under certain conditions. Yet parallel work has documented real risks: Fan, Tang, and colleagues found in the British Journal of Educational Technology that students using generative AI showed signs of what they called metacognitive laziness, delegating the monitoring and regulation of their own learning processes to the tool, with measurable effects on motivation, learning processes, and performance. Studies of student perceptions, including a widely cited survey by Stöhr, Ou, and Malmström covering students across genders, academic levels, and fields of study, showed that usage patterns and perceptions of AI chatbots vary systematically by discipline and demographic characteristics. Jensen, Buhl, Sharma, and Bearman, reviewing the claims made in the first months after ChatGPT’s release, documented how rapidly academic discourse swung from speculative enthusiasm to alarm, often without empirical grounding.

Institutional adoption, the new analysis emphasizes, is not simply a matter of individual choice. Research grounded in established adoption frameworks, such as the Technology-Organization-Environment model applied by Jiang, Wei, Qiu, and Huang to Chinese higher education, shows that organizational structures, regulatory environments, and competitive pressures shape diffusion as much as user attitudes do. A study by Luo, Zhou, and Cui of Chinese universities and technical and vocational colleges examined generative AI adoption among faculty specifically, while work by Kim, Klopfer, and colleagues documented divergent perceptions of generative AI between faculty and students in university courses, a gap that complicates any single-institution policy. Historically, faculty adoption of learning technologies has been slow and uneven; Liu, Geertshuis, and Grainger’s systematic review of academics’ adoption of learning technologies, and Porter and Graham’s study of institutional drivers and barriers to blended learning, both identified workload, incentive structures, and institutional support as decisive factors. A recent book by James Hutson, The Adoption of Artificial Intelligence and Inertia in Higher Education: Exploring Complex Resistance to Technological Change, published by Routledge, frames this resistance as structurally rational rather than merely conservative, an argument the review engages directly.

China presents one of the most aggressive state-driven adoption trajectories in the world. Following the emergence of DeepSeek, a domestically developed large language model, Chinese universities launched dedicated courses on the technology in early 2025, as reported by Reuters, seeking to capitalize on the domestic AI boom. The Ministry of Education has publicly addressed AI in education policy statements, and research by Liu on regulations, technology policies, and universities’ attitudes toward artificial intelligence in China documents a coordinated national posture. Scholars including Luo, Zhou, and Cui, and earlier work on the diffusion of AI in Chinese higher education, suggest that the combination of ministerial direction, funding, and competitive ranking pressure has accelerated adoption in ways that decentralized systems, particularly in North America and much of Europe, have struggled to replicate. Meanwhile, the theoretical framing of what “global” higher education means has itself been contested, with Marginson’s work on the geopolitics of higher education and Yang and Tian’s rethinking of the global through the Chinese concept of tianxia offering alternatives to methodological nationalism in higher education research, an approach Shahjahan and Kezar critiqued as early as 2013.

The picture in Africa and other majority-world regions is starkly different, shaped less by questions of pedagogy than by infrastructure, access, and historical inequities. Systematic reviews of generative AI in African higher education document both opportunities, including expanded tutoring capacity and multilingual support, and serious ethical concerns around data sovereignty, colonial patterns in AI systems, and uneven preparedness. Studies of AI literacy in Ghanaian and Nigerian universities mapped affective, behavioral, cognitive, and ethical dimensions of students’ relationships to the technology. The critical scholarship is even sharper: Birhane’s analysis of algorithmic colonization of Africa, Posada’s work on the coloniality of data work in Latin America, and Zembylas’s decolonial strategies for undoing digital neocolonialism in teaching and learning all point to the uncomfortable fact that the global AI economy depends heavily on low-paid data workers in the Global South, a dynamic exposed most famously by reporting that OpenAI used Kenyan workers earning less than two dollars per hour to make ChatGPT less toxic. Research from South Africa on curriculum transformation and decolonization, and on AI adoption’s benefits and ethical implications across African institutions, raises the question of whether AI in these contexts functions as aid or obstacle.

Latin America shows a similar pattern of rapid uptake constrained by structural inequality. The Digital Education Council’s 2026 survey of AI in higher education across Latin America and studies measuring perceived academic skills associated with ChatGPT use among Latin American students document substantial student engagement with the tools, often ahead of institutional policy. UNESCO’s International Institute for Higher Education in Latin America and the Caribbean has pushed for accelerated digital transformation across the region. But as in Africa, the risk identified by scholars is that universities in these regions become consumers of technologies built elsewhere, trained on data extracted from their populations, governed by policies designed for different contexts. Williamson and Komljenovic’s analysis of edtech investors’ techno-financial “futuring” of higher education adds a political economy dimension: venture capital is actively shaping what the university of the future is imagined to be, and institutions that lack resources to build or even critically evaluate these systems may find their priorities increasingly set externally.

Policy responses are converging slowly and unevenly. Jin, Yan, Echeverria, Gašević, and Martinez-Maldonado analyzed institutional adoption policies and guidelines across a global sample and found wide variation in rigor, specificity, and enforcement, with many institutions settling for vague principles rather than operational guidance. The Stanford Institute for Human-Centered Artificial Intelligence’s 2026 AI Index Report tracks the accelerating capability of frontier models, and the United Nations Development Programme’s 2025 Human Development Report frames the choice before societies as a matter of how AI reshapes human possibility rather than whether. Within the research community, O’Dea has asked whether generative AI constitutes a genuine paradigm shift for higher education, and Bearman, Ryan, and Ajjawi traced the discourses through which AI has been constructed in the field, warning that language itself shapes what institutions believe is possible or inevitable.

The synthesis offered by the Michigan State analysis lands on a cautionary note that resonates across all the regional literatures it surveys. The evidence does not support a simple verdict that generative AI either enhances or degrades learning; rather, outcomes depend on design, scaffolding, and the metacognitive demands placed on learners, as Yan, Greiff, Teuber, and Gašević argued in Nature Human Behaviour when mapping the promises and challenges of generative AI for human learning. A systematic review by Long, Wang, Md Rashid, and Lu found that AI’s impact on student engagement is mediated by teaching methods, and Mokoena and colleagues’ review of AI in higher education institutions cataloged functionalities, challenges, and best practices that point toward intentional rather than reactive adoption. The deeper risk, on this reading, is not that universities will adopt AI too quickly, but that they will adopt it thoughtlessly, importing vendor-driven solutions in a seller’s market of the kind Teräs and colleagues described in the post-COVID education technology landscape, without the decolonial, pedagogical, and political scrutiny that the global evidence now demands.

What emerges from the global comparison is that the future of generative AI in higher education will not be decided by the technology itself but by the institutional, national, and geopolitical choices surrounding it. Wealthy institutions in North America and Europe are debating academic integrity policies while their counterparts in Accra, Lagos, São Paulo, and Nairobi confront the same tools with fewer resources and different historical burdens. China is integrating AI into curricula at ministerial speed. Students everywhere are already using these tools regardless of policy, as the survey evidence consistently shows. The task scholars increasingly articulate is to replace the default narratives of inevitability and resistance with empirical, globally aware analysis of what these systems actually do to learning, to academic labor, and to the distribution of educational power, before the decisions that matter most have been made by default.