1.  Introduction

AI is becoming a defining force in education, but its adoption is uneven. OECD data show that students in high-income countries utilize AI for learning at a rate more than three times higher than that in low-income regions [1]. During COVID-19 school closures, over 463 million children had no access to remote learning [2]. These disparities underscore that technology alone cannot guarantee equal opportunity. Existing studies often focus on ethical debates or advanced applications in wealthy contexts, while affordable and scalable approaches (including offline tools, synchronous classrooms, and open educational resources, OER) have received insufficient attention [3,4]. A systematic, cross-regional analysis of how such practices contribute to educational equity remains limited.

This paper aims to address that gap by applying a three-dimensional equity framework that considers opportunity, process, and outcomes. Guided by this framework, the central research questions are: How can AI-driven initiatives be effectively designed to expand opportunities for marginalized learners? In what ways do these practices foster fairness in learning processes? To what extent can such interventions reduce disparities in educational outcomes across regions?

By formulating these questions, the study seeks to contribute to ongoing debates about the role of AI in education. Rather than assuming that technology will inherently exacerbate or alleviate inequality, this research examines the conditions under which AI can cultivate more inclusive and equitable learning environments.

2.  Technological empowerment: innovations breaking resource barriers

The first pathway to achieving equity involves overcoming structural barriers. Affordable offline devices and cross-regional sharing systems show how inclusive design can extend access to underserved learners. These technologies represent not merely technical innovations but also novel avenues of opportunity, ensuring that students who were previously excluded gain a chance to participate in structured learning.

2.1.  Offline learning technologies

Offline devices such as solar-powered tablets and preloaded learning kits, ensure educational continuity in contexts with limited connectivity. Studies has confirmed their resilience, particularlyin crisis contexts [1]. In sub-Saharan Africa, these devices have reached over half a million students, leading to improvements in mathematics and literacy outcomes without internet dependence. Cost-effectiveness strengthens their appeal. Television-based learning in India was demonstrated to cost only one-twentieth of smart classrooms, while still delivering significant learning gains [5,6]. Such initiatives demonstrate that affordability and scale can coexist.

Apart from the results that can be measured by data, offline technology has also had a considerable social impact. According to the reports given by teachers, in rural areas of Kenya, students who used adaptive tablets showed higher participation in class, and parents also found that their children's enthusiasm for studying at home has increased significantly compared to before. These feedback situations indicate that fairness is not merely about making hardware devices accessible to everyone; the relevant learning process is also crucial. By reducing reliance on urban-centered infrastructure, offline solutions directly promote opportunity equity. As a result, rural schools can be in line with national standards.

2.2.  Cross-regional resource sharing

Fairness is actually also related to the distribution of high-quality resources among different regions. In China, synchronous classrooms connect rural schools with elite schools in cities, thus creating opportunities for students in remote areas to study advanced courses [7]. Internationally, UNESCO is promoting open educational resources. Regard it as a way to narrow the gap, ensure that resources are universally accessible and have relatively strong adaptability [4]. The World Bank has also found that countries that have adopted a shared resource model have benefited a wider population during the pandemic [3] .

Beyond national boundaries, the Open Educational Resources Cooperation Platform in Latin America and Africa has illustrated how cross-regional knowledge exchange promotes procedural equity. Spanish resources developed in Chile have been adapted and then used in rural schools in Bolivia. The initiative launched by the African Union is to promote the development of multilingual open educational resources to support the literacy of English and local languages. This approach not only enhances people's access to learning materials but also cultivates learners' sense of belonging and cultural inclusiveness. However, obstacles still exist The dialect and accent bias existing in speech recognition poses a risk of excluding learners [8]. If adjustments are not made seriously, the resource sharing platform may make the existing inequality even more serious.

3.  Teaching transformation: toward equitable personalization

AI not only broadens access to content but also transforms the modalities of teaching and learning, enabling greater personalization. Although personalization has often been criticized as privileging already advantaged learners, inclusive applications demonstrate its potential to advance equity. When implemented carefully, AI systems can provide that those with learning barriers receive targeted support, while simultaneously uncovering opportunities for those whose potential might otherwise remain untapped.

3.1.  Support for special groups

One of the most tangible impacts of AI on equity is its ability to support learners with disabilities and learning difficulties. Adaptive reading platforms adjust text complexity and provide real-time scaffolding, have proven effective for students with dyslexia or other reading impairments. Similarly, text-to-speech and speech-to-text technologies mitigate barriers for learners with visual or auditory challenges, facilitating more active classroom participation [4].

International practical examples can illustrate the extent of this transformation. Take the United States for instance, artificial intelligence reading assistants can improve the comprehension scores of children with dyslexia by adjusting their vocabulary and reading speed. Looking at Kenya, solar-powered devices with integrated voice recognition functions have been installed. Enable students with impaired hearing to keep up with the pace of the class in real time. From these measures, it can be known that artificial intelligence tools can adjust teaching methods based on the needs of learners themselves, rather than requiring learners to follow that inflexibility system. In this way, the disadvantages of the system can be alleviated.

Although there are some controversies, AI-driven emotion recognition systems still show some hope in supporting students' mental health. In classrooms with relatively limited resources, teachers are confronted with a high student-to-teacher ratio. Artificial intelligence can help identify learners who show signs of detachment or distress. If detected in time, teachers can intervene before students fall far behind, thereby reducing the risk of students dropping out of school. The United Nations Children's Fund (UNICEF) [2] emphasizes that this inclusive technology is crucial for ensuring fairness in marginalized communities, where students often have limited access to counseling and personalized support.

However, these applications also carry risks. If the algorithms misunderstand facial expressions, gestures or language patterns, it can lead to incorrect identification of students, those from minority cultural or language groups [8]. Just as research shows, facial recognition technology often does not perform well in analyzing the expressions of black or Asian students. This may not alleviate stigmatization but instead intensify it. These challenges highlight the significance of designing culturally sensitive algorithms and maintaining teacher supervision. Personalized technology should not replace educators. Instead, it should be used as an auxiliary tool to complement professional judgment.lization technologies should serve as supportive tools, complementing professional judgment.

3.2.  Identifying potential and expanding opportunities

Personalization can rely on identifying potential capabilities that standardized education systems might overlook to expand educational equity. Intelligent tutoring systems adjust teaching content based on learners' learning paces and provide personalized learning approaches to help learners master knowledge. VanLehn [9] found through research that The role played by adaptive systems can be comparable to that of human tutoring, which indicates that adaptive systems have great potential to tap into the abilities of students who may fall behind.

The relevant evidence from Brazil clearly demonstrates all the transformative roles of adaptive platforms in expanding educational opportunities. Geekie is a widely used adaptive education tool that enables students in low-income communities to access pre-university resources that were once only available to students from elite educational institutions [10]. By using this tool, these students not only improved their test scores but also gained the opportunity to enter higher education. This is a very crucial outcome in a region where there is a serious inequality in education.

The large-scale adaptive projects carried out in India are another example of an effective approach, which is to integrate adaptive modules into TV-based communication. During the COVID-19 pandemic, these projects identified outstanding students in rural communities and provided them with scholarship opportunities [5,6]. By integrating personalization into accessible forms, these initiatives have also given students outside urban centers more access to education.

Artificial intelligence also helps vocational education and workforce development. In Southeast Asia, governments of various countries have launched pilot projects of AI-driven skills assessment to identify the strengths of rural youth and recommend training programs that match emerging industries such as digital services and renewable energy. These systems have narrowed the gap between academic performance and employability. Expand the scope of fairness from the school stage to the lifelong learning stage.

From these examples, it can be seen that personalization has great transformative potential in expanding access to resources and increasing opportunities. However, whether personalization can achieve good results depends on its design and implementation. When systems are trained on relatively narrow datasets, they may misclassify learners and overlook their true potential. The accent bias existing in speech recognition [8] indicates that students using non-standard dialects may not receive accurate support. If the dataset is not diverse enough and the system does not continuously evaluate, personalization may not be able to identify hidden talents, and instead, existing stereotypes may persist. To enable personalization to realize its potential for promoting equality, it is crucial to ensure that it can reliably identify and expand opportunities.

3.3.  Equitable data and ethical design for personalization

The third crucial aspect of fair personalization is to design a fair and inclusive data governance system. Artificial intelligence models rely on the data used to train them. If the training data is biased, it is inevitable that biased results will be obtained. In an educational environment, decisions regarding tracking, evaluation, and opportunities will have a profound impact on life opportunities. The risks involved here are particularly high.

When speech recognition systems handle non-mainstream dialects, they often perform poorly, which leads to a higher error rate and is more likely to cause misunderstandings [8] . Martin and Wright found that the language of African Americans was seriously misread, which shows how the bias hidden in the training data marginalizes the entire group. Similarly, other studies have also documented significant errors when dealing with regional English variants.

Fair personalization requires proactive strategies to address these biases. The dataset should be expanded to include different languages, dialects, and cultural backgrounds, and a participatory design approach involving teachers, students, and local communities should be adopted. To ensure that personalized tools can reflect users' actual life situations rather than some abstract assumptions. The framework of moral governance should require that the algorithmic decision-making process be more transparent, allowing educators and policymakers to supervise and correct unfair phenomena. Additionally, moral personalization needs to balance the relationship between data collection and privacy rights. Excessive intrusive monitoring, such as continuous emotional monitoring, may lead to a classroom environment characterized by distrust. International organizations like UNESCO [4] have called for the establishment of a prudent governance framework that respects both fairness and the dignity of students, combines personalization with human rights, and ensures that learners receive support rather than merely being treated as data points.

Personalized explanations achieved through artificial intelligence indicate that fairness does not mean providing the same resources for everyone. On the contrary, fairness demands that teaching be tailored to the needs, strengths and potential of learners. Inclusive personalization addresses the obstacles faced by disabled learners, broadens the paths for disadvantaged students, and ensures that opportunities are not predetermined by an individual's socioeconomic background. Fair personalization requires strong safeguard measures. For instance, culturally sensitive design, inclusive datasets, and transparent governance. When designed and implemented responsibly, AI-driven personalization can achieve fairness in the process and results. It enables learners to succeed not because they are different, but because their differences are recognized and valued. By transforming classrooms into Spaces that can accommodate diversity, Artificial intelligence personalization caan realize a broader vision of education as a public good, that is, making education accessible to all, inclusive and empowering for all.

4.  Collaborative ecosystems: shared responsibility for equity

Technological tools alone cannot guarantee educational equity. Sustainable equity requires ecosystems in which governments, companies, educators, and international organizations collaborate to share responsibility. AI initiatives are most effective when embedded within broader systems of accountability and adaptation.

4.1.  Corporate social responsibility

Private companies play a pivotal role in scaling educational technologies. Google’s free tools in Brazil expanded access to high-quality resources in marginalized communities [10]. Public–private partnerships in Africa, combining corporate technology with government infrastructure, have also demonstrated effectiveness [1].

Tensions persist, as commercial priorities may exclude disadvantaged groups. The “GenAI divide” identified by Beckman et al [11]. illustrates that advanced generative tools reach wealthier students while low-income peers are excluded. Clearer governance frameworks are needed to align corporate strategies with equity goals.

4.2.  Teacher innovation and professional networks

Teachers are central to the creation of equitable systems.. AI enhances learning only when educators effectively adapt and integrate it into their pedagogical practices. Collaborative networks allow teachers to share materials, co-develop lessons, and adapt resources across cultures. UNESCO highlights teacher participation as essential for the sustainable use of OER use [4].

Brazil’s adaptive platforms succeeded partly through teacher collaboration on student data [10]. In China, synchronous classrooms thrived when rural and urban teachers co-facilitated lessons [7]. These examples demonstrate that equity depends as much on empowering teachers as on distributing technology.

4.3.  International cooperation and policy frameworks

International organizations provide crucial coordination and governance. UNESCO’s OER Recommendation calls for global sharing and sustainable policies [4]. The World Bank found that countries adopting collaborative approaches achieved wider reach during the pandemic [3].

Equity also depends on addressing algorithmic bias. Studies on speech recognition indicate that the uncritical adoption of AI systems risks excluding marginalized learners [8]. Policies promoting inclusive datasets, local language integration, and algorithmic transparency are essential for achieving equitable outcomes.

4.4.  Toward sustainable ecosystems

Sustainable ecosystems require aligned contributions: governments provide infrastructure, companies drive innovation, educators adapt practices, and global organizations coordinate. Embedding initiatives within national strategies, such as India’s low-cost television education [5,6], ensures continuity beyond donor funding cycles. Teacher training embedded in curriculum reforms also strengthens resilience.

Collaborative ecosystems demonstrate that equity is a systemic issue rather than a purely technological one. AI tools achieve their potential only when supported by responsible companies, empowered teachers, and coordinated international governance. Shared responsibility transforms fragmented projects into sustainable strategies for achieving global educational equity.

5.  Conclusion

This paper has examined how artificial intelligence can serve as a lever for advancing global educational equity through affordable technologies, inclusive personalization, and collaborative ecosystems. By applying the three-dimensional equity framework—opportunity, process, and outcomes—it has shown that AI is not inherently a force of inequality but can, under the right conditions, promote educational fairness.

An analysis of offline learning technologies shows that low-cost and scalable tools such as solar devices and TV-based teaching can significantly expand access opportunities in less serviced environments. Initiatives for cross-regional resource sharing, including synchronous classrooms and open educational resources, have explored how to promote equity by redistributing specialized knowledge. Personalized learning technology highlights the potential of artificial intelligence in supporting disadvantaged learners and also demonstrates its latent capabilities, thus expanding the opportunities for achieving process and outcome fairness. The collaborative ecosystem involving governments, enterprises, educators, and international organizations indicates that achieving sustainable equity requires the joint responsibility of multiple parties.

However, there are still significant challenges. Algorithm-based biases in speech recognition systems may marginalize learners from non-dominant language or cultural groups. The "genAI gap" still excludes low-income students from cutting-edge tools, which has raised concerns about a new layer of inequality. Sustainability relies on incorporating low-cost innovations into national policies rather than relying on projects driven by donors.

Future research should focus on three aspects: conducting a systematic assessment of inclusive artificial intelligence in different cultural contexts, which is crucial for ensuring adaptability. The governance framework needs to address the inherent ethical and fairness issues in algorithm design and deployment, and longitudinal studies are required to evaluate the long-term impact of AI-supported fairness initiatives on learning trajectories and social mobility.

Overall, if the development and deployment of artificial intelligence are guided by the principles of inclusiveness, affordability and collaboration, it can make meaningful contributions to global educational equity, establish a sustainable ecosystem, and combine technological innovation with human institutions. This is crucial for transforming fragmented projects into lasting and equitable strategies.