AI in Mathematics Classrooms: A Synergistic Strategy for Diversity and Inclusivity

Synthesis: Jita, Jita, and Omoniyi synthesize research on artificial intelligence in Math Education and ask a narrower question than most reviews of the topic: under what conditions does AI make a mathematics classroom more inclusive rather than merely more efficient? Anchored in Universal Design for Learning and the SAMR Model, they conduct a hybrid conceptual–structured narrative review of 38 sources published between 2019 and 2025, drawn from 213 records in Scopus, Web of Science, and ERIC and coded with substantial agreement (Fleiss' κ = 0.75). It catalogues adaptive platforms, Intelligent Tutoring systems, and multilingual multimodal tools, then weighs them against infrastructure, bias, and cost. Its distinctive contribution is the Synergistic AI-Inclusive Mathematics Education (SAIME) framework, which treats UDL as the inclusion-oriented design foundation and SAMR as a technological-transformation pathway, with teacher-mediated Pedagogies and Teaching Strategies and ethical safeguards as conditions rather than add-ons. The governing claim is that a higher SAMR level does not guarantee Inclusive Learning: redefinition can still exclude students whose language, devices, or connectivity do not fit the design.

Key Findings

1. The review identified 213 records (107 Scopus, 65 Web of Science, 41 ERIC), removed 58 duplicates, and synthesized 38 sources published 2019–2025 after excluding 30 at full-text appraisal.
2. Adaptive platforms and intelligent tutoring systems were the most consistently reported affordances, adjusting content, difficulty, and pacing to students' real-time performance in mixed-ability classrooms.
3. SAMR level and inclusion were treated as independent dimensions: an application can redefine a mathematical task while remaining inaccessible because of linguistic, Accessibility, or infrastructural barriers.
4. Case evidence was sparse but specific: Squirrel AI in China, chatbot tutors piloted in under-resourced African settings, and Philippine tutoring platforms linked to gains in understanding and confidence.
5. Barriers clustered in four areas: unreliable infrastructure, teachers' limited AI training, algorithmic bias in training data, and recurring licensing, hosting, and security costs.
6. SAIME organizes integration into input/design (UDL and SAMR), process (AI functionalities plus teacher mediation), and output (access, participation, achievement), with ethical safeguards operating across all three layers.

Framing inclusion as a design problem

The paper treats mathematics as a discipline where exclusion is structural rather than incidental. Students with disabilities, linguistic minorities, and low-income learners face barriers that inflexible teaching methods do not absorb, and AI integration without deliberate design can reproduce the disparities it promises to remove.

Two frameworks carry the argument. Universal design for learning, developed by David Rose and Anne Meyer at CAST in the late 1980s and 1990s and adapted from Ronald Mace's architectural concept, supplies three tenets: multiple means of engagement, representation, and action and expression. The SAMR model, formulated by Puentedura in 2010, supplies a progression from substitution and augmentation to modification and redefinition.

The two are not interchangeable: UDL sets the pedagogical direction, SAMR grades the depth of technological change. The authors also note each framework's limits — UDL assumes more capable adaptive technology than currently exists, and SAMR's linear staging oversimplifies recursive experimentation.

What the synthesis says AI can do

Three functionality clusters dominate the findings. Adaptive platforms adjust content, difficulty, and pacing against real-time performance, which the authors present as the most direct route to Personalized Learning in mixed-ability classrooms. Intelligent tutoring systems supply step-by-step guidance, hints, and misconception repair for students who receive little individual attention otherwise. Language processing and multimodal tools translate explanations across languages and render mathematical ideas as speech, visuals, or braille.

Affordances are grouped pedagogically: inclusion support, differentiation, reduced mathematics anxiety, real-time Feedback and Formative Assessment, collaborative grouping through intelligent pairing, and teacher-facing analytics. The evidence is often small-scale, but the cases are concrete: Squirrel AI in China, chatbot tutors in African contexts where teacher access is limited, and Philippine tutoring platforms associated with gains in understanding and confidence among students struggling with abstract concepts.

Where the evidence stops

The constraints the review identifies are not technical footnotes. Uneven device provision, unreliable electricity, and unstable connectivity mean that under-resourced schools cannot use the tools at all, which converts an inclusion technology into a mechanism for widening the Digital Divide. Teacher preparation is a second bottleneck: many teachers lack training and confidence to integrate AI, and some fear losing human interaction.

Algorithmic bias is the third: because systems learn from historical data, predictive analytics can misjudge students with disabilities or minority-language speakers, so bias mitigation, transparency, and co-design are preconditions rather than post-deployment fixes. Cost closes the list: licensing, hosting, and security are recurring expenses low-income institutions struggle to sustain.

The SAIME framework

SAIME is a cyclical, three-layer model. The input/design layer pairs UDL's three means with SAMR's four levels; the process layer connects AI functionalities to teacher-mediated strategies — differentiated instruction, collaborative learning, culturally sustaining pedagogy, responsive intervention and Scaffolding; the output layer specifies equitable access, participation, and achievement. Ethical and equity safeguards run across all three layers, with a feedback loop tying outcomes back into redesign.

The framework's analytical move is to deny that transformation and inclusion are the same axis. An application can redefine a task at the highest SAMR level while remaining inaccessible to a student who cannot read the interface language or lacks a device. Equity thus becomes the criterion for judging whether an AI-enabled transformation has pedagogical value, and the Teaching stays central: analytics inform judgment, they do not replace it.

What this means for practice

• Instructors. Audit accessibility gaps with UDL checklists before selecting a tool, then judge tools by the SAMR level they reach, not by vendor claims.
• Instructors. Keep interpretation of AI analytics in professional hands; the framework assigns diagnosis and response to teachers, not to dashboards.
• Curriculum designers. Map each AI-supported task against all three UDL means, since a tool that differentiates difficulty may offer only one mode of representation.
• Administrators. Budget for connectivity, devices, assistive technology, and multilingual functionality alongside licenses, because the review treats infrastructure as a precondition for equitable use.

Limitations

• The corpus is 38 sources published between 2019 and 2025 and drawn from three databases, so the themes reflect what those indexes surface.
• The design is a hybrid conceptual–structured narrative review without primary data, effect sizes, or inferential statistics; SAIME is explicitly a preliminary framework awaiting empirical validation.
• The authors acknowledge possible publication bias, since positive results are more likely to be published and therefore to shape the themes.

Citation

Jita, T., Jita, L. C., & Omoniyi, A. A. (2026). AI in mathematics classrooms: A synergistic strategy for diversity and inclusivity. Cogent Education, 13(1).