Artificial Intelligence in Education: Applications and Limitations for Teachers in Low-and Middle-Income Countries

Miguel Landa-Blanco^*

• National Autonomous University of Honduras, Tegucigalpa, Honduras

The advent of generative artificial intelligence (AI)-advanced machine learning systems capable of autonomously producing coherent, contextually relevant text, imagery, and multimedia content (Hashmi & Bal, 2024) -has rapidly transformed numerous sectors, including education (Memarian & Doleck, 2023;Mohd Amin et al., 2025). However, the available literature on AI largely neglects low-and middle-income countries (LMICs), with scholarly attention predominantly directed toward advanced economies. Thus, data from AI adoption in LMICs is limited (Khan et al., 2024). In LMICs, where educational systems are frequently constrained by resource limitations, infrastructure deficits, and pedagogical challenges (Betthäuser et al., 2023;Delprato & Antequera, 2021;Little, 2006;Naparan & Alinsug, 2021), generative AI offers an unprecedented opportunity to reimagine teaching and learning (Díaz & Nussbaum, 2024). Yet, this technological promise is accompanied by complex socio-technical and ethical dilemmas that require careful navigation if the potential benefits are to be realized equitably and sustainably (Adel et al., 2024). This paper provides a brief overview of the pedagogical applications of AI for teachers in LMICs, with a particular focus on how these tools can both support and potentially constrain educational outcomes. Building on these considerations, this paper argues that while generative AI offers meaningful pedagogical benefits for teachers in resource-constrained settings, including improved linguistic access, support for instructional design, and more effective differentiation, its equitable adoption in LMICs ultimately depends on resolving key infrastructural, contextual, and governance challenges. Expanding Educational Access: Overcoming the Language Barrier Scientific discourse is predominantly disseminated through English-language publications (Márquez & Porras, 2020). The hegemony of English in global scientific discourse constitutes a profound structural barrier for scholars in LMICs (Ramírez-Castañeda, 2020), impacting both the production and consumption of knowledge. On the production side, limited access to advanced English-language education and editorial resources disproportionately hinders researchers' ability to contribute to highimpact publications (Di Bitetti & Ferreras, 2017), secure international funding, and participate in transnational academic networks, conditions that reinforce systemic disparities in scholarly visibility and influence. Equally consequential is the effect on scientific consumption: the overwhelming prevalence of English-language literature restricts the accessibility of cutting-edge research for non-English speaking practitioners, educators, and policymakers (Amano et al., 2016;Arenas-Castro et al., 2024), thus affecting resource availability for instruction. This linguistic asymmetry not only curtails the integration of global evidence into local contexts but also perpetuates epistemic hierarchies that marginalize alternative knowledge systems and local epistemologies. In this way, the monolingual orientation of contemporary science operates as a mechanism of exclusion, entrenching global inequities in the circulation, recognition, and application of knowledge (Polanco & Mayorga, 2025; "Scientific Publishing Has a Language Problem," 2023).However, advances in AI offer promising avenues for mitigating the linguistic barriers that have long constrained the equitable participation of LMICs scholars in global scientific discourse (Salani & Tapfuma, 2025). High-quality, context-sensitive translation tools powered by AI can facilitate both the production and consumption of academic knowledge across linguistic divides (Gulati et al., 2024), enabling researchers and teachers to draft, revise, and disseminate manuscripts in English with greater precision and efficiency, while also enhancing access to English-language publications for non-Anglophone readers. The employment of AI tools to enhance grammatical accuracy, improve textual readability, and facilitate effective language translation is generally regarded as an ethically appropriate practice in scholarly environments (Cheng et al., 2025). Moreover, AI-driven platforms can support real-time multilingual collaboration. Such capacity enables the localization of curricula, fostering inclusivity and engagement by embedding cultural relevance and context within learning materials. A recent systematic review synthesizes empirical evidence on the pedagogical affordances of AI for teachers, identifying key advantages across three interrelated domains: instructional planning, classroom implementation, and assessment (Celik et al., 2022). In the planning phase, AI systems facilitates data-informed decision-making by providing insights into students' socio-academic backgrounds and supporting the selection of instructional content based on readability or learning needs. During implementation, AI technologies enable real-time monitoring of student engagement and cognitive states, allowing for timely, adaptive interventions that not only enhance instructional responsiveness but also support personalized learning. These systems can also enrich the instructional experience by promoting more dynamic teacher-student interactions and fostering greater instructional enjoyment. In the domain of assessment, AI contributes to automating evaluative processes, such as essay scoring and plagiarism detection, arguably improving both the efficiency and objectivity of student evaluation. The automation of labor-intensive tasks, such as grading and lesson plan generation (Gurl et al., 2025;Jukiewicz, 2024;Li et al., 2025;Liu et al., 2022), reallocates teachers' time toward more meaningful pedagogical engagement, individualized instruction, and mentorship.AI can simplify complex concepts by adapting explanations to match students' varying age groups and comprehension levels. Through advanced natural language processing and pedagogical modeling, AI can break down intricate ideas into more accessible content based on lexical, syntactical, and discourse-level simplification (Smirnova et al., 2025). This customization often involves generating simplifications that distill key information into digestible portions, as well as crafting analogies that relate new concepts to familiar experiences or everyday objects. AI can be used to tailor such analogies based on the student's cultural context and cognitive demands (Cao et al., 2024). Thus, AI supports differentiated and inclusive learning (Kooli & Chakraoui, 2025), making challenging material more engaging and understandable for diverse student populations, thereby enhancing overall educational effectiveness. Despite its transformative potential, the deployment of generative AI in LMICs education systems is shaped and often constrained by persistent infrastructural and systemic barriers (Farooqi et al., 2024). Access to reliable electricity, stable internet connectivity, and affordable digital devices remains uneven, especially in rural and marginalized communities. Without substantial investment in digital infrastructure, AI-driven educational interventions risk exacerbating existing inequities rather than alleviating them (Khan et al., 2024).Additionally, generative AI models are fundamentally data-driven, relying on extensive corpora predominantly sourced from Western contexts and major world languages. This epistemic foundation introduces the risk of reproducing and amplifying cultural biases (Foka & Griffin, 2024;Tao et al., 2024). In LMICs, where cultural heterogeneity and historical legacies of colonialism shape social realities, AI-generated materials that lack local contextualization may inadvertently perpetuate stereotypes, misrepresent histories, or convey inappropriate normative assumptions (Ferrara, 2023;van Kolfschooten & Pilottin, 2024).In this sense, AI-generated content may also be imprecise, distorted, or entirely fabricated, posing risks to the accuracy and integrity of academic work (Sun et al., 2024). As such, both teachers and students should approach AI outputs critically (Ng et al., 2021), not as definitive sources, but as preliminary drafts or scaffolds that require careful evaluation, verification, and refinement (Salido et al., 2025). Integrating AI into the learning process should thus emphasize human oversight and intellectual engagement, rather than passive dependence (Yue Yim, 2024). Therefore, rather than treating AI-generated content as authoritative, learners should be taught to interrogate, verify, and revise AI outputs, using them as springboards for deeper inquiry rather than endpoints. This pedagogical stance not only guards against misinformation and bias but also reinforces essential cognitive skills such as evaluation, synthesis, and argumentation. As AI becomes increasingly embedded in classrooms and academic workflows, it is imperative to reframe its role, not as a substitute for human judgment, but as a dynamic partner in the co-construction of knowledge (Kohnke et al., 2025). For LMICs, this shift presents a unique opportunity: to build more inclusive, adaptable, and resilient education systems that are both technologically empowered and pedagogically grounded.From a pedagogical standpoint, overreliance on AI-generated content and responses could attenuate critical thinking and deep cognitive engagement among learners if not carefully scaffolded (Gerlich, 2025;Zhai et al., 2024). In this sense, a recent study concluded that using AI for essay writing led to reduced brain connectivity, lower cognitive engagement, and weaker performance compared to using search engines or writing without tools (Kosmyna et al., 2025). Compounding this issue, many public educational institutions in LMICs may lack both clear policies and the technological infrastructure needed to detect AI-generated content. Without institutional guidelines or access to reliable detection tools, teachers are left to rely on intuition or inconsistent methods to identify work produced by AI, which can lead to confusion, mistrust, and uneven enforcement of academic integrity. This regulatory gap not only undermines fair assessment practices but also makes it more challenging to cultivate a shared understanding among students and educators about the appropriate and ethical use of AI in learning environments (Azevedo et al., 2024;Chan, 2023). In the absence of proactive policy frameworks, AI use in classrooms risks evolving in ad hoc, unmonitored ways that may further entrench existing educational inequities. Ultimately, generative AI should be understood not as a panacea but as a pedagogical aid, one that requires thoughtful integration, sustained teacher training, and robust human oversight to ensure its benefits are equitably and ethically realized (Kohnke et al., 2025;Wiese et al., 2025). The integration of generative AI into educational systems presents a critical challenge for low-and middle-income countries: how can this technology be meaningfully adopted in contexts marked by limited infrastructure, scarce funding, and uneven digital literacy? Although AI holds promise for instructional planning, language accessibility, and classroom support, its broader implications for educators require scrutiny. Generative tools can personalize learning (Merino-Campos, 2025), translate complex content, and increase teacher efficiency (Tan et al., 2025), advantages that are particularly appealing in resource-constrained environments. However, these benefits must be weighed against the socio-technical, infrastructural, and epistemic conditions that shape educational practice. Without serious efforts to localize content, mitigate algorithmic bias, and invest in digital infrastructure, AI may exacerbate rather than alleviate existing inequalities. AI should no longer be treated as a peripheral innovation (Ma et al., 2025); its responsible integration requires deliberate strategies that center both educators and learners.At the same time, educators in these contexts face long-standing structural and economic challenges . Teachers often work under intense pressure, managing overcrowded classrooms, limited materials, and competing demands on their time (Delprato & Antequera, 2021;Little, 2006;Naparan & Alinsug, 2021). The arrival of AI does not automatically reduce these burdens; in fact, it often introduces new technical and ethical complexities (Eyal, 2025;Nguyen et al., 2023). For AI to be usable and beneficial, teachers need more than digital tools; they need training that supports their ability to engage critically and confidently with AI in the classroom. Building AI literacy requires targeted professional development (Pei et al., 2025). This involves not only technical instruction but also the cultivation of ethical and contextual awareness so that educators can evaluate and adapt AI for their specific teaching environments (Gouseti et al., 2025). To understand what such integration demands in practice, it is helpful to examine the concrete conditions under which many schools in LMICs operate.In this sense, the case of Honduras may serve to exemplify a broader reality in many LMICs, where schools face severe infrastructural challenges that hinder both learning environments and the effective adoption of educational technologies. The Honduran public education system currently lacks a formal regulatory framework for the use of AI. Yet, there are promising early efforts underway. Initiatives such as teacher symposiums and short-term training programs aim to empower educators and promote the responsible integration of AI into the classroom (Secretaría de Educación de Honduras [Honduran Ministry of Education], 2023). However, 60% of schools in the country require roof repairs, 50% have damaged walls, and 12% do not have electricity. Water supply is often irregular, and many schools lack proper sanitation infrastructure. Only half have sinks, 70% of toilets need repair, and 9% of schools do not have waste disposal systems. Furniture shortages are also widespread, with 60% lacking sufficient chairs (FONAC, 2024). A similar pattern is evident in Sub-Saharan Africa, where studies highlight that limited electrification, deteriorated school buildings, and shortages of basic furniture constrain both teaching quality and student learning (Hassan et al., 2022). These physical deficits severely limit the feasibility of implementing digital tools. Without reliable electricity, safe facilities, and adequate learning environments, the adoption of AI risks remaining a distant aspiration rather than a practical solution. Any meaningful implementation must be accompanied by investment in foundational infrastructure to ensure that technology addresses, rather than deepens, educational inequality.In conclusion, AI must be reimagined as a tool for equity (Garcia Ramos & Wilson-Kennedy, 2024; Kohnke & Zaugg, 2025), not just efficiency. This means positioning teachers as co-designers and informed decision makers, rather than passive adopters. Policies should prioritize locally developed, adapted, or curated AI systems (Hsu et al., 2022), training programs that build critical digital capacity, and implementation strategies rooted in the material and cultural realities of schools in low and middle-income countries. Beyond technical training, educators should be supported to question whose knowledge is embedded in AI systems, and whose is omitted. These questions are essential to prevent AI from reinforcing marginalization. The goal is not to replicate the technological trajectories of high-income countries, but to carve out educational futures that are contextually responsive, culturally grounded, and led by those who understand the classroom from within. For teachers in LMICs, this means participating actively in shaping technological change, not just responding to it. However, LMICs are not a uniform group; their economic, social, and political conditions differ significantly, leading to varied educational challenges and outcomes (Local Burden of Disease Educational Attainment Collaborators, 2020). Effective policies must therefore be tailored to each country's specific context and developmental priorities. A coherent policy agenda for LMICs should focus on three interconnected priorities. First, investment in foundational infrastructure is essential, since reliable electricity, safe facilities, and adequate learning environments determine whether AI tools can be used at all. Second, sustained teacher training is needed to build both technical competence and the critical capacity to evaluate and adapt AI for local pedagogical purposes. Third, clear governance frameworks must guide procurement, data use, and algorithmic accountability so that AI adoption aligns with national educational goals and avoids reinforcing existing inequities.Generative AI's role in education must also be examined through the lens of systemic readiness and institutional governance. Sustainable integration in LMICs depends on policy coherence, ethical regulation, and the establishment of data governance frameworks that ensure privacy and accountability (Papagiannidis et al., 2025). Governments must prioritize building AI governance capacities aligned with educational goals, including developing national strategies to guide procurement, data use, and algorithmic transparency. At the institutional level, universities and teacher-training colleges can function as key incubators for AI literacy, embedding critical and context-sensitive use of AI tools into teacher education programs (Daher, 2025;Kelley & Wenzel, 2025). Equally important is fostering South-South collaboration, allowing LMICs to share regionally relevant best practices and locally developed tools rather than relying solely on imported technologies shaped by external pedagogical assumptions. When aligned with inclusive policy design and institutional capacity building, AI can move beyond pilot interventions to become an embedded component of equitable educational reform, strengthening local ownership and long-term sustainability (Khan et al., 2024). • Dog Breeds: All dogs are related, but they look different. Small changes make big differences over time.• El cuello largo de la jirafa: Las jirafas con cuellos más largos podían alcanzar más hojas. Con el tiempo, más jirafas tenían cuellos largos.• Razas de perros: Todos los perros están relacionados, pero se ven diferentes. Pequeños cambios hacen grandes diferencias con el tiempo.6th Grade (11-12 years)• Peppered Moths: In England, trees got dark from pollution. Dark moths survived better than light ones.• Finches on Islands: Finches on different islands had different beaks, depending on what food was there.• Polillas moteadas: En Inglaterra, los árboles se oscurecieron por la contaminación. Las polillas oscuras sobrevivieron mejor que las claras.• Pinzones en las islas: Los pinzones de diferentes islas tenían picos distintos, según el tipo de alimento.9th Grade (14-15 years)• Darwin's Finches: Beak differences helped finches eat different foods. This is adaptation in action.• Antibiotic Resistance: Some bacteria survive medicine and multiply. That's evolution happening now.• Los pinzones de Darwin: Las diferencias en los picos ayudaron a los pinzones a comer diferentes alimentos. Esto es adaptación en acción.• Resistencia a los antibióticos: Algunas bacterias sobreviven a los medicamentos y se multiplican. Eso es evolución en tiempo real.12th Grade (17-18 years)• Hardy-Weinberg as a Recipe: Like a recipe, allele frequencies stay the same unless ingredients change (e.g., mutation, migration).• Human-Primate DNA: Humans share ~98% of DNA with chimps-strong molecular evidence of shared ancestry.• Hardy-Weinberg como receta: Como una receta, las frecuencias génicas se mantienen igual a menos que cambien los "ingredientes" (mutaciones, migración).• ADN humano y de primates: Los humanos comparten ~98% de su ADN con los chimpancés-una fuerte evidencia molecular de ancestros comunes.Note. AI-generated example using ChatGPT (4.1-mini) using the prompt: "I am a teacher planning lessons on the theory of evolution for 3rd, 6th, 9th, and 12th graders. Please provide specific, detailed analogies and engaging stories tailored for each grade level to help students understand key concepts of evolution. Use simple language for younger students and include more scientific depth for older students. Focus on examples that illustrate natural selection, adaptation, variation, and speciation. For each grade, list two clear analogies or stories that make the ideas concrete and relatable. Present the output in a table. Include a Spanish translation of the content".