Status Quo

Though AI integration in education – for teachers, students, and lifelong learners – in the Global South remains uneven, a number of regional case studies may provide inspiration. In Latin America, children in K–12 schools have helped shape training data for non-English, region-specific AI systems. In India, partnerships with the Mozilla Foundation have expanded access to emerging technologies and context-specific learning tools in rural communities. In Indonesia, similar initiatives aim to narrow educational and technological divides between rural and urban regions. However, access to AI remains limited in many rural areas, where digital connectivity, education, and institutional capacity are often insufficient or contingent on infrastructure controlled by regional or Global North partners.

At the same time, students increasingly turn to AI tools for tutoring, advice, and social interaction, raising concerns about overreliance, misinformation, and the displacement of human support. Mainstream LLMs such as ChatGPT expand access to information but remain prone to generating inaccurate content. As culturally informed models are created, AI can become a lifelong tool for continued learning.

Inciting Event(s)

Despite the rapid spread of Ed-Tech platforms, educational outcomes across many Global South contexts remain stagnant because AI tools are often introduced without adequate teacher support, local content, infrastructure, or evidence of pedagogical value. Uncritical or poorly supervised use of AI contributes to student disengagement, weaker independent problem-solving, and fewer opportunities for peer interaction. Teachers and students alike feel increasingly disconnected from one another. Policymakers and educators recognize that the solution is not removing AI from schools but integrating it more thoughtfully into traditional learning environments.

A groundbreaking study challenges the assumption that more technology automatically leads to better learning. Instead, it finds that poor educational outcomes stem largely from misguided implementation of both traditional and AI-enabled Ed-Tech systems, highlighting especially severe disparities in the Global South. The study’s findings prompt education ministries, teacher organizations, and parent groups to demand that public procurement and school deployment meet common standards for learning quality, safety, and inclusion. In parallel, philanthropic networks launch advocacy campaigns emphasizing the benefits of AI in education when deployed responsibly and contextually. The campaigns culminate in an international summit for the development of a global mechanism for AI in education, resulting in a South-South Regulatory Framework for AI in Education.

Mechanism(s)

The South-South cooperative framework for AI in education covers a range of needs and requires meaningful human oversight: Teachers and schools retain responsibility for assessment, learner support, and high-stakes decisions, while AI tools provide transparent, adaptable assistance rather than automated judgment.

This assistance includes, for example, mobile-first systems for resource-scarce communities and efficient low-resource models, and technical best practices for developers and implementers in the education space, such as age verification, transparent UI supporting AI literacy, screen-time limits, and safeguards restricting AI engagement to educational purposes in learning environments. The framework also provides normative guidance around broader issues like psychological profiling, emotional dependency, and peer isolation.

The framework’s combination of high-level guidance and technical tooling gives participating countries access to shared technical resources, policy guidance, and tested implementation models to support policy decisions and identify solutions already tailored to differences across Global South contexts – whether regional, linguistic, urban or rural, socioeconomic, etc. For example, public schools are able to tailor coursework to students’ individual interests and learning styles while ensuring that it also remains age-appropriate and developmentally grounded. As adoption expands, localized data collection and deployment become priorities in rural areas to ensure educational materials reflect local languages, cultures, and realities. Over time, personalized learning systems evolve alongside students, adapting to their intellectual growth and changing interests.

Beyond K–12 education, new upskilling pathways emerge for university graduates and young adults looking to use AI for entrepreneurship and professional development.  Self-directed learning initiatives help aspiring entrepreneurs build AI-powered businesses that respond to local economic needs.

Outcomes

In this scenario, context-specific AI-enabled Ed-Tech solutions emerge through collaboration between local universities, foundations, and major technology companies. These efforts create AI tutoring systems capable of reaching students in remote areas with lessons adapted to local languages and cultural contexts, bridging gaps that traditional infrastructure fails to address. Through an international cooperative framework led by Global South governments but with buy-in from global AI leaders, Global South stakeholders also establish guardrails that protect students from psychological profiling and unhealthy emotional dependence on AI, ensuring technology supports learners without exploitation.

As a result of the standards set by this international framework, the educational landscape is transformed for the better in the Global South. In classrooms, AI takes over repetitive instruction and administrative burdens, allowing teachers to focus on mentorship, collaborative project-based learning, and meaningful one-on-one engagement that nurtures creativity and critical thinking. Outside of K-12 education, this shift contributes to a more educated and adaptable workforce. With younger generations gravitating increasingly towards entrepreneurial and self-directed careers, AI-powered tools make it easier to learn new skills on demand, diversify sources of income, and achieve greater flexibility than traditional employment models. The result is a more inclusive learning economy, in which quality educational and entrepreneurial opportunities are less constrained by geography and income, while sustained public investment helps address remaining barriers to access.

Next Steps: Achieving This Scenario by 2031

Alternative scenarios are, first and foremost, thought exercises that illustrate different versions of potential futures. The next phase of the Responsible AI Fellowship will build policy and technical playbooks to make these desired futures a reality by 2031. Key to this process is the solicitation of suggestions from partners, community members, and regional experts to identify practical tools and pathways; if you have ideas for how particular parts of this scenario can become reality, please fill out the survey here:

Stimson Responsible AI Fellowship Scenario Feedback – Fill out form

In the meantime, we’ve drafted several potential mechanisms to spark discussion, which are included below.

Policy Options

Creating Interoperable Child Safety Tools

A key dilemma in age gating software amid child safety concerns is interoperability — smaller companies may struggle to meet age gating requirements across a variety of jurisdictions, particularly as different countries approach liability and responsibility for age gating. The joint framework can address this problem by establishing shared international norms for liability in age gating among countries in the Global South, along with an interoperable set of technical requirements that ease compliance burdens for smaller operators in the Global South.

Joint Pilots for Ed-Tech in the Global South

In developing the framework, the coalition can identify potential technical solutions to common problems — such as using mobile-first tech to address digital infrastructure gaps — and deploy them in test environments across the Global South to gather a diverse range of data. These pilots can be used to evaluate possible standards for inclusion in the global framework. Each pilot should include independent evaluation of learning outcomes, accessibility, privacy, teacher workload, and student wellbeing, with feedback from learners, educators, caregivers, and local authorities.

Relevant Facts

Regional Case Studies

Latin America

The Stereotypes and Discrimination in Artificial Intelligence (EDIA) toolkit comprises 35,000 sentences authored by students to help train Latam-GPT data.

South and Southeast Asia

The Responsible Computing Challenge (RCC) helps to promote computer science education in both rural and marginalized communities across India. In Indonesia, Universitas Terbuka partnered with Microsoft Azure AI to enhance asynchronous tutoring opportunities for a growing student population, boosting performance.

Internet, AI Connectivity and Education

Share of Population Using the Internet, 2025

Fellowship countries have rates as low as 31.2% (Tanzania) in contrast to 93.1% of the United States.

Estimated Share of Working-age Adults Who Use Generative AI, March, 2026

Representative of population aged 15-64 who have used generative AI. Fellowship countries have rates as low as 7.6% (Tanzania) in contrast to 31% of the United States.

AI Integration into Education

International Cooperation

The Inter-American Development Bank (IDB) Education Division partnered with South Korean researchers on a comprehensive report of AI-enabled EdTech solutions in Korea and lessons for Latin America and the Caribbean (LAC). In 2021, UNESCO released a report titled AI and education: Guidance for policy-makers as part of the Global Education 2030 Agenda. Policy recommendations include developing system-wide readiness and cost-value assessments; establishing and monitoring measurable targets to ensure inclusion, diversity, and equality in AI development; and steering AI development and policy towards the protection of human rights and gender equality.

Overreliance and Skills Atrophy

In a preprinted (not yet peer-reviewed study), students using large language models (LLM) to complete essays experience 55% less Dynamic Direct Transfer Function (DDTF) than those who use only their brain; on average those using search engines experience 34-48% less.

• 83% of LLM users reported difficulty in quoting their essays.
• 94.44% of students using only their brain claimed full ownership of their essays, whereas students using LLM claimed only partial credit between 50-90%.
• This study suggests that overreliance on generative AI models contributes to a loss of deep memory encoding processes and cognitive agency.