Feminist and Global South perspectives on AI-supported learning environments

Artificial intelligence in education (AIEd) is being promoted and introduced as a transformative solution to persistent education challenges, such as teacher shortages, large class sizes, and uneven education quality. Yet, as it is designed and used, AIEd is not and can not deliver on its promise for Global South learners, especially with the existing digital divide and access-related issues. Without fundamental changes, AIEd risks reproducing the same gendered, economic, linguistic, and geopolitical inequalities that already shape education and that it claims to address, particularly in the Global South.

A feminist and Global South perspective asks whose knowledge, needs, languages, and futures are being centered in AIEd’s design and use, apart from how AI can improve learning. These technologies—learner-facing tools such as adaptive learning platforms and tutoring systems; teacher-facing tools for assessment, feedback, and administrative tasks; and institution-facing systems that analyze patterns such as enrollment, retention, and student progress—carry embedded assumptions that, if left unchallenged, could reproduce the challenges that already exist. Girls, women, rural learners, low-income students, and linguistically marginalized communities may be positioned as future beneficiaries of AIEd while remaining underrepresented in its design, governance, and implementation.

The mismatch between rapid AIEd expansion within structurally unequal, gendered, and context-specific realities in the Global South limits its ability to deliver on its promise. We map four structural tensions through which this mismatch plays out before proposing what it means for the Global South to lead an equitable AIEd transformation:

1. Access vs. meaningful participation

The Global South continues to face a layered digital divide, where unequal access intersects with gender, geography, and socio-economic status. Only 27% of individuals in low-income countries are connected to the internet, compared to over 93% in high-income countries, reflecting persistent structural gaps. Marginalized groups are disproportionately excluded from AI benefits, and gender inequality compounds this exclusion. In low-income countries, for example, only 10% of girls and young women (aged 15-24) are online, compared to 22% of boys and men in the same age group.

Importantly, access does not always translate to meaningful participation in many Global South contexts, especially in South Asia, where girls may have physical access to devices but limited autonomy in using them, which restricts sustained engagement with AI tools and widens the AI literacy gap.

2. Contextual relevance vs. epistemic injustice

Many AI systems are designed in Global North contexts and exported around the world. Such “solutionist” AI approaches, where technology is assumed to solve complex social problems, risk failure when they overlook local pedagogies, teacher capacity, and infrastructure. AI learning platforms often lack local language support, making them less usable in multilingual regions. This can create dependency without meaningful learning integration, especially for under-resourced schools.

AI systems frequently prioritize dominant languages and knowledge systems, marginalizing indigenous or local knowledge, reinforcing global hierarchies of knowledge production. This increases the risk of enforced homogenization and a narrowing of what counts as valid knowledge, pedagogy, and learning caused by how these systems are trained.

3. Gender inclusion vs. bias amplification

Women are less likely to access and use digital technologies across the Global South. Research looking at a variety of indicators of digital adoption found the greatest gender gaps concentrated within Africa, with large sub-national disparities also occurring in India and Pakistan.

But even when AI programs target inclusion, structural barriers, including norms around mobility, time use, safety, and financial autonomy, limit women’s participation. Women are underrepresented  not just as users but as creators, designers, and decision makers in AI systems.. The absence of women from the design of AIEd, reinforces gender bias embedded in datasets and systems. As a result, AI holds the potential to personalize learning, but biased datasets may reinforce stereotypes (e.g., career suggestions aligning with gender norms rather than learner aspirations).

4. Adoption vs. constrained readiness

The adoption of AI often takes place within uneven conditions of readiness. The reluctance toward generative AI in developing countries, frequently read as technophobia, is often a response to real structural constraints, including limited infrastructure, gaps in digital literacy, weak governance systems, and concerns about cultural relevance. In this sense, AI integration is shaped by both principled resistance, ethical, pedagogical, and epistemic concerns, and constrained resistance arising from limited resources and institutional capacity. In such situations, AIEd risks becoming a superficial add-on rather than a meaningful tool for educational transformation.

Implications for policy and practice

AIEd requires attention to meaningful participation, contextual relevance, epistemic and linguistic justice, institutional readiness, and ethical governance. To address these tensions, governments, AI developers and funders must collaborate to:

• Bridge access and meaningful participation: Governments need to put in place strategic frameworks that will enable sustainable bridging of the infrastructure and inclusivity gap. This could be achieved by providing digital infrastructure through the education and schooling system. Governments should also work toward ensuring that the implementation of community infrastructure is gender responsive, by, for example, ensuring that last-mile digital rural community centers are owned and/or managed by women, and enforcing gender responsiveness as a standard in AI system design.
• Invest in local knowledge, languages and governance: Governments should invest in local-language datasets, support local knowledge in AI training, and fund research that can build models. Local co-design and community ownership should as well be part and parcel of the solution being provided. For these two approaches to succeed, there is further need to enhance AI skill sets among locals, including teachers, administrators, and local researchers, through well-structured AIEd capacity development initiatives to make this sustainable. Lastly, AIEd adoption must be regulated by local governments and embedded within existing infrastructures to support their audit and adaptation to fit local learning needs.
• Redesign AI systems with and for women and girls: Given the low level of women’s participation in AI ecosystems, AI developers need to shift their focus on a deliberate attempt to redesign AI models and datasets with the aim of ensuring a shift from male-centric design of AIEd systems and closing the gender gap. A key element to achieving this is to ensure that women have access to technical AI jobs and that they are able to retain them once they secure them, as well as mandatory bias audits of AIEd models and datasets, with reporting of gender outcomes.
• Condition funding on equity standards and infrastructure support: Funders promoting AIEd should make sure that the conditions under which this technology is introduced are equitable. This means that investment decisions should be conditioned on evidence of community participation, gender-responsive design, local readiness and capacity. Funders should also invest in broader enabling conditions for equitable AIEd, including research capacity and strengthening of local datasets.

Conclusion

AI has and will continue to offer enormous potential within the education sector, but that potential runs parallel with the risks that may be associated with AIEd. As we have argued, the barriers for learners in the Global South are structural, gendered, and epistemic. Addressing them requires governments and policymakers to invest in local and equitable infrastructure, knowledge, and governance and to recognize that for AI in education to work for all learners, the Global South must position itself as a partner in its design and use.