Aynur Abdullayeva
Senior Lecturer
Department of Humanities
Azerbaijan State Academy of Physical Education and Sport
Azerbaijan, Baku
ORCID: 0009-0007-9003-0236<\/p>\n\n\n\n
Keywords:<\/strong> arti\ufb01cial intelligence, inclusive education, special educational needs, assistive technology, adaptive learning, Universal Design for Learning, educational equity<\/p>\n\n\n\n Abstract<\/strong>. Inclusive education, grounded in the principle that all learners — regardless of ability, background, or learning pro\ufb01le — should have equitable access to quality instruction, has become a central policy priority worldwide. Concurrently, arti\ufb01cial intelligence (AI) technologies are increasingly integrated into classrooms through adaptive learning platforms, intelligent tutoring systems, assistive communication tools, and automated assessment applications. This article presents a systematic review of empirical and theoretical literature examining the role of AI in advancing inclusive education. Following a structured search of major academic databases and a thematic synthesis of the retrieved studies, in line with PRISMA reporting principles, four principal domains of AI application are identi\ufb01ed: (1) personalized and adaptive instruction, (2) assistive technologies for students with disabilities, (3) support for teachers through data-driven decision-making, and (4) inclusion of linguistically and culturally diverse learners. The review also highlights persistent challenges, including algorithmic bias, unequal access to digital infrastructure, insu\ufb03cient teacher preparation, and data privacy concerns. Drawing on the Universal Design for Learning framework and existing typologies of AI in education, the article proposes policy and practice recommendations for the responsible integration of AI into inclusive educational systems. The \ufb01ndings suggest that while AI holds considerable promise for reducing barriers to learning, its bene\ufb01ts are contingent on deliberate, equity- oriented design and implementation.<\/p>\n\n\n\n Introduction<\/strong><\/p>\n\n\n\n Inclusive education has been enshrined as a global policy goal since the 1994 Salamanca Statement [11] and rea\ufb03rmed through Sustainable Development Goal 4, which calls for inclusive and equitable quality education for all learners [12]. In practice, however, education systems continue to struggle to accommodate the full diversity of student needs, including learners with disabilities, students from linguistic and cultural minorities, and those who require differentiated pacing or modality of instruction. Over the past decade, arti\ufb01cial intelligence (AI) has moved from a peripheral experimental technology to a mainstream feature of educational software, encompassing intelligent tutoring systems, speech and text recognition tools, automated captioning, adaptive assessment engines, and generative AI tutors.<\/p>\n\n\n\n The convergence of these two trends \u2014 the policy imperative of inclusion and the rapid diffusion of AI in classrooms \u2014 raises an important question: to what extent, and under what conditions, can AI technologies genuinely advance inclusive education rather than reproduce or deepen existing inequities? This article addresses that question through a systematic review of the academic literature published between approximately 2018 and 2025. The review is guided by three<\/p>\n\n\n\n research questions: (RQ1) What categories of AI application are reported in the literature as supporting inclusive education? (RQ2) What learning and participation outcomes are associated with these applications? (RQ3) What barriers and risks are identi\ufb01ed that may limit or undermine the inclusive potential of AI?<\/p>\n\n\n\n Main part.<\/strong> Inclusive education is commonly understood not merely as the physical placement of students with diverse needs in mainstream classrooms, but as a systemic commitment to removing barriers to participation and achievement for all learners. The Universal Design for Learning (UDL) framework, developed by Meyer, Rose, and Gordon [8], offers a widely used pedagogical lens for this commitment. UDL proposes that curricula should provide multiple means of engagement, representation, and action\/expression, so that variability among learners is treated as the norm rather than the exception. This framework is frequently invoked in the AI-in-education literature as a benchmark against which the inclusivity of a given technology can be assessed.<\/p>\n\n\n\n Ouyang and Jiao [9] propose a widely cited typology distinguishing three paradigms of AI in education: AI-directed learning, in which the system controls the pace and content and the learner is a recipient; AI-supported learning, in which the system scaffolds learner-centered activities while preserving learner agency; and AI-empowered learning, in which AI augments the learner’s and teacher’s own capabilities through collaborative, symbiotic interaction. This typology is useful for analyzing the literature on inclusive education, since different AI applications occupy different positions along this spectrum, with implications for learner autonomy and agency.<\/p>\n\n\n\n A growing body of scholarship cautions that AI systems can encode and amplify existing biases present in their training data, with disproportionate effects on already marginalized groups. Baker and Hawn [1] provide a comprehensive account of algorithmic bias in education, documenting how predictive and adaptive systems can systematically disadvantage students based on race, language background, or disability status. Fu and Weng [5] similarly call for human-centered, responsible AI practices in education, arguing that ethical safeguards must be embedded in the design process rather than added retrospectively. These concerns are directly relevant to inclusive education, where the populations most likely to bene\ufb01t from AI support are often also those most vulnerable to algorithmic harm.<\/p>\n\n\n\n Following identi\ufb01cation and screening of titles and abstracts, full texts of potentially eligible studies were assessed against the eligibility criteria. Included studies were analyzed using thematic synthesis, in which recurring concepts were coded and grouped into higher-order themes. This process yielded the four thematic domains presented in Section 4.1\u20134.4, together with a cross-cutting set of barriers presented in Section 4.5. As a review of secondary literature rather than a primary data-collection study, this article does not report new statistical \ufb01ndings; instead, it aims to consolidate and critically synthesize existing empirical evidence in order to inform theory, policy, and practice.<\/p>\n\n\n\n The most frequently reported application of AI in the reviewed literature concerns personalization: adaptive learning platforms and intelligent tutoring systems that adjust content di\ufb03culty, pacing, and feedback according to an individual learner’s demonstrated performance. Ouyang and Jiao [9] note that such systems allow instruction to be differentiated at a scale that would be di\ufb03cult for a single teacher to achieve unaided. For learners with attention di\ufb03culties, processing differences, or gaps in prior knowledge, adaptive sequencing has been reported to reduce frustration and support sustained engagement, though the reviewed studies caution that personalization algorithms trained predominantly on majority-population data may perform less reliably for atypical learning pro\ufb01les.<\/p>\n\n\n\n A second major theme concerns AI-enabled assistive technologies designed explicitly for students with disabilities. Barua et al. [2] review a range of AI-enabled personalized assistive tools developed for children with neurodevelopmental disorders, including autism spectrum condition and attention-de\ufb01cit\/hyperactivity disorder, spanning speech-generating devices, emotion-recognition applications, and adaptive communication boards. Hopcan et al. [7] similarly document a systematic body of research on AI in special education, reporting applications in automatic speech recognition for students with speech and language impairments, real-time captioning for deaf and hard-of-hearing learners, and computer-vision-based tools supporting students with visual impairments. Chalkiadakis et al. [3] extend this discussion to the combination of AI with virtual reality, reporting that immersive, AI-adapted environments can support both academic accessibility and social inclusion for students with disabilities by allowing controlled, repeatable practice of real- world scenarios.<\/p>\n\n\n\n A third theme concerns the role of AI in supporting, rather than replacing, teachers. Garg and Sharma [6] report that AI-based tools can assist special-needs educators by automating routine assessment and documentation tasks, freeing time for direct instruction, and by providing data-driven insights into individual student progress that inform differentiated planning. Several reviewed studies emphasize that AI functions most effectively as a decision-support tool that augments professional judgment, rather than as a substitute for teacher expertise, particularly in contexts requiring nuanced understanding of a student’s social, emotional, and behavioral needs.<\/p>\n\n\n\n A fourth theme, somewhat distinct from disability-focused applications, concerns the use of AI to support students from linguistic and cultural minority backgrounds. Salas-Pilco et al. [10] conduct a systematic review of AI and related technologies used to support minority students, identifying applications such as machine-translation-supported instruction, culturally responsive adaptive content, and AI-mediated communication that reduce language barriers between students, families, and schools. Fitas [4] similarly reports on AI applications that simultaneously address special-needs support and language-barrier reduction, arguing that these two dimensions of inclusion are often addressed by overlapping technological approaches, such as multimodal AI systems capable of translation, simpli\ufb01cation, and adaptive scaffolding within a single platform.<\/p>\n\n\n\n Across the reviewed literature, four recurring barriers are identi\ufb01ed. First, algorithmic bias: Baker and Hawn [1] document cases in which predictive models embedded in adaptive systems produced systematically less accurate or less favorable outputs for students from historically marginalized groups, underscoring the need for bias auditing prior to deployment. Second, unequal access to digital infrastructure: several studies note that the bene\ufb01ts of AI-enabled inclusion are contingent on reliable internet connectivity, appropriate devices, and technical support, which are unevenly distributed within and across countries, risking a widening rather than narrowing of educational gaps. Third, insu\ufb03cient teacher preparation: multiple reviewed studies report that teachers frequently lack training in how to interpret AI-generated recommendations, evaluate their appropriateness for a given student, and integrate them into existing individualized education plans. Fourth, data privacy and consent: the collection of granular behavioral and, in some cases, biometric data from students with disabilities raises distinct ethical concerns regarding consent, data ownership, and the potential for stigmatizing pro\ufb01ling, a concern echoed by Fu and Weng [5] in their call for human- centered, responsible AI governance in educational settings.<\/p>\n\n\n\n Read together, the reviewed literature suggests that AI’s contribution to inclusive education is best understood not as a uniform effect but as a set of distinct application types, each with different implications for learner agency and equity. Mapped onto the typology proposed by Ouyang and Jiao [9], assistive technologies for students with disabilities often function in an AI-supported or AI-empowered mode, augmenting the learner’s own communicative or cognitive capacities while preserving agency; by contrast, some adaptive-sequencing systems risk functioning in a more AI-directed mode, in which the system’s determination of a learner’s pathway constrains rather than expands choice. This distinction matters for inclusive education speci\ufb01cally, because the UDL principle of providing multiple means of engagement and expression is more fully realized when learners retain meaningful agency over how they engage with adapted content, rather than being passively routed through it.<\/p>\n\n\n\n The discussion also highlights a tension between individualization and standardization. AI systems designed for scale often rely on data patterns drawn from majority populations, which can inadvertently disadvantage learners whose needs or response patterns diverge from those norms — precisely the learners that inclusive education aims to serve. Addressing this tension requires deliberate design choices, including diverse and representative training data, participatory design<\/p>\n\n\n\n involving students with disabilities and their families, and ongoing bias auditing, consistent with the responsible — AI principles articulated by Fu and Weng [5].<\/p>\n\n\n\n Conclusions<\/strong><\/p>\n\n\n\n This systematic review indicates that arti\ufb01cial intelligence holds meaningful potential to advance inclusive education through personalized instruction, assistive technologies for students with disabilities, decision-support tools for teachers, and applications that reduce language and cultural barriers. At the same time, this potential is not automatic: it depends on deliberate, equity-oriented design, implementation, and governance. Based on the synthesized evidence, six recommendations are proposed for policymakers, school leaders, and developers: <\/p>\n\n\n\n 1) embed Universal Design for Learning principles into the design and procurement of educational AI tools;<\/p>\n\n\n\n 2) require bias auditing and disaggregated performance reporting for AI tools used in inclusive settings;<\/p>\n\n\n\n 3) invest in structured professional development that equips teachers to critically interpret AI-generated recommendations;<\/p>\n\n\n\n 4) address infrastructural inequities directly through investment in connectivity, devices, and technical support in under-resourced schools;<\/p>\n\n\n\n 5) establish clear, context-appropriate data governance and consent frameworks for behavioral and biometric data from students with disabilities; and 6) support participatory design processes that involve students with disabilities, families, and special education professionals in the development and evaluation of AI tools intended for inclusive settings. Future research should prioritize longitudinal, mixed-methods studies that examine not only academic outcomes but also learner agency, social inclusion, and long-term equity effects of AI adoption in diverse educational systems, including comparative studies across countries with differing levels of digital infrastructure.<\/p>\n\n\n\n References<\/strong><\/p>\n\n\n\n <\/p>\n","protected":false},"excerpt":{"rendered":" Aynur AbdullayevaSenior LecturerDepartment of HumanitiesAzerbaijan State Academy of Physical Education and SportAzerbaijan, BakuORCID: 0009-0007-9003-0236 Keywords: arti\ufb01cial intelligence, inclusive education, special educational needs, assistive technology, adaptive… <\/p>\n","protected":false},"author":1,"featured_media":1349,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[34],"tags":[],"class_list":["post-1353","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ie"],"_links":{"self":[{"href":"https:\/\/journal.pedaqoq.az\/index.php?rest_route=\/wp\/v2\/posts\/1353","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/journal.pedaqoq.az\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/journal.pedaqoq.az\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/journal.pedaqoq.az\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/journal.pedaqoq.az\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1353"}],"version-history":[{"count":1,"href":"https:\/\/journal.pedaqoq.az\/index.php?rest_route=\/wp\/v2\/posts\/1353\/revisions"}],"predecessor-version":[{"id":1354,"href":"https:\/\/journal.pedaqoq.az\/index.php?rest_route=\/wp\/v2\/posts\/1353\/revisions\/1354"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/journal.pedaqoq.az\/index.php?rest_route=\/wp\/v2\/media\/1349"}],"wp:attachment":[{"href":"https:\/\/journal.pedaqoq.az\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1353"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/journal.pedaqoq.az\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1353"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/journal.pedaqoq.az\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1353"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}\n