Artificial Intelligence in Mathematics Education

ASES EDUSCI ASES INTERNATIONAL JOURNAL OF EDUCATION SCIENCE ISSN 2822-6844 Year (2025) Vol:5 Issue: 2 Issued in December 2025 www.e-edusci.com 671 Article Arrival Date Article Published Date 30.11.2025 20.12.2025 Artificial Intelligence in Mathematics Education Bekir DANIŞ Dr., Aydın Adnan Menderes University, Digital Transformation Office, Orcid:0000-0002-3532-2670 Abstract This article comprehensively examines the current development, theoretical foundations, pedagogical implications, and applications of artificial intelligence (AI) methods in mathematics education. In recent years, advances in deep learning, natural language processing, generative models, and student modeling techniques have led to transformative innovations in mathematics education, such as adaptive learning, automated problem solving, personalized feedback, and pedagogical agents. This study discusses the epistemological impacts of AI on mathematical thinking processes and assesses critical limitations such as algorithmic biases, data privacy, verifiability, pedagogical fit, and cognitive dependency. The findings suggest that AI offers significant opportunities in mathematics education but requires careful pedagogical, ethical, and methodological considerations. Keywords: Mathematics, education, artificial intelligence 1. Introduction Artificial intelligence has become a growing focus of academic interest within educational technologies since the 2010s. Mathematics education is positioned as a critical area for AI applications because it requires abstract thinking, symbolic manipulation, reasoning processes, and problem-solving skills (Holmes et al., 2019). The emergence of generative models such as ChatGPT, DeepMind AlphaMath, and Khanmigo has created a paradigmatic shift in mathematics education through the automation of mathematical reasoning processes, modeling of student performance, and the expansion of personalized instruction opportunities (Kasneci et al., 2023). The use of artificial intelligence in the context of mathematics education is not limited to the application of new technologies; it also raises fundamental questions about the nature of learning processes. In particular, the extent to which AI-enabled systems can support higherlevel cognitive processes such as mathematical thinking, conceptual understanding, and abstract reasoning is a topic of increasing debate in the existing literature. Systems that recognize students' error patterns, suggest solution strategies, or model the problem-solving process offer new possibilities in instructional design, but they also necessitate a reinterpretation of the place of mathematical knowledge in cognitive development. This necessitates a theoretical framework for balancing the pedagogical roles of AI with the teacher's directive/inquiry guidance. ASES EDUSCI ASES INTERNATIONAL JOURNAL OF EDUCATION SCIENCE ISSN 2822-6844 Year (2025) Vol:5 Issue: 2 Issued in December 2025 www.e-edusci.com 672 Furthermore, the methodological diversity of AI research in mathematics education is also noteworthy. Current studies span a wide spectrum, from explainable artificial intelligence (XAI)-based student modeling approaches to predictive performance models based on largescale learning analytics data and verification mechanisms that assess the quality of generative AI solutions. However, the literature lacks a holistic assessment of the extent to which these different methods align with the epistemological and didactic principles of mathematics education. Therefore, future research should focus on theoretically grounded, interdisciplinary methodologies that clarify the relationship between algorithmic approaches and pedagogical goals. This article systematically examines the theoretical foundations, application areas, and research agenda of AI technologies in mathematics education. 2. Theoretical Framework Artificial intelligence-based educational technologies have become increasingly sophisticated in terms of modeling, monitoring, and personalizing learning processes. AI systems used in education generally fall into four main categories. Adaptive Learning Systems create personalized learning experiences by monitoring student performance in real time and adjusting content delivery accordingly (Liu & Koedinger, 2017). Intelligent Tutoring Systems (ITS), on the other hand, provide instant, targeted feedback by modeling the student's cognitive state and optimize the learning process (VanLehn, 2006). Educational Analytics and Student Modeling, another field complementing these systems, focuses on predicting student behavior, learning patterns, and success probabilities using machine learning methods. Generative Learning Models, which have gained prominence in recent years, can automate high-level cognitive processes such as problem solving, explanation generation, and content creation using large language models (ChatGPT, Gemini, Claude, etc.). Mathematical cognition consists of multidimensional processes that form the basis of learning. These include skills such as conceptual understanding, computational fluency, logical reasoning, representational transformations (e.g., translating verbal expression into symbolic representation or graphical information into algebraic form), and strategy selection. Modern AI systems, particularly with the development of large language models, have become capable of automatically processing many of these cognitive processes. AI models can provide direct cognitive support to the learning process through their capacity to perform representational transformations, analyze patterns of student errors, derive process-oriented problem-solving steps, and generate mathematical explanations. From this perspective, AI is no longer merely a tool for evaluating learning outcomes; it has become a complementary and transformative component of learning processes. These AI-based systems draw on diverse theoretical approaches to understanding and modeling mathematical cognition. For example, cognitive load theory explains how optimal learning occurs within the limits of a learner's working memory capacity, and AI systems can apply this theory to offer adaptive guidance to reduce cognitive load when students experience difficulty. Similarly, constructivist learning theories emphasize the processes by which students actively construct mathematical meaning, while AI-enabled tools can support this process by providing students with opportunities to experiment, reflect, and reconstruct their own solution strategies. In this context, AI is not only a technological component that models cognitive processes but also a pedagogical tool that can be translated into concrete classroom applications of various learning theories. ASES EDUSCI ASES INTERNATIONAL JOURNAL OF EDUCATION SCIENCE ISSN 2822-6844 Year (2025) Vol:5 Issue: 2 Issued in December 2025 www.e-edusci.com 673 Additionally, the Explainable AI (XAI) approach is becoming increasingly important in mathematics education. Because mathematical accuracy requires high precision, a transparent understanding of why and how an AI model generates a solution is critical for both pedagogical trust and evaluation processes. XAI techniques can enhance teacher-student-AI interactions by deconstructing the solution steps suggested by the model, identifying the source of errors, or providing insight into students' faulty reasoning patterns. This presents a new research paradigm in mathematics education that requires AI-based systems to produce solutions that are not only performance-driven but also pedagogically interpretable and didactically meaningful. 3. Applications of Artificial Intelligence in Mathematics Education 3.1. Automated Problem Solving and Solution Explanation Large language models have evolved to generate solutions for complex algebra, calculus, and probability problems, while offering detailed, step-by-step explanations (Trinh et al., 2024). This capability significantly contributes to students' ability to compare various solution strategies and enhances their self-regulated learning skills by providing immediate, granular insight into the problem-solving process. 3.2. Personalized Mathematics Instruction In the realm of personalized instruction, AI-based adaptive platforms play a crucial role by instantly measuring a student's knowledge level, dynamically adjusting the difficulty of tasks, and predicting potential learning barriers (Koedinger & Aleven, 2016). A prominent example of this approach is Khan Academy's AI-based tool, "Khanmigo," which utilizes these mechanisms to tailor the educational experience to individual needs. 3.3. Automated Assessment AI-supported assessment and evaluation systems have advanced to the point where they can analyze open-ended mathematical expressions, classify specific errors within a student's solution path, and seamlessly integrate with symbolic algebra systems. Recent studies indicate that deep learning algorithms can now classify student solutions with an accuracy that rivals human evaluation (Piech et al., 2015), offering scalable and precise feedback mechanisms. 3.4. AI-Supported Reasoning in Mathematical Proofs Automated theorem provers such as Lean, Coq, and Isabelle are increasingly being utilized for educational purposes to support reasoning in mathematical proofs. Through these tools, students can verify their proof steps with AI assistance, explore alternative proof strategies, and instantly identify logical errors in their work, thereby fostering a deeper understanding of mathematical logic. 3.5. Mathematical Content Creation with Generative Models Generative models assist mathematics teachers in content creation by automating several key administrative and pedagogical tasks. These AI tools enable educators to efficiently generate new questions, prepare student-specific worksheets, and present a variety of solution methods, effectively reducing workload while enriching instructional materials (Kasneci et al., 2023). 4. Opportunities AI-based educational approaches significantly enhance the quality of learning environments by providing diverse opportunities for learning mathematics. The first of these opportunities is the potential to reduce learning inequalities. Personalized learning systems help narrow ASES EDUSCI ASES INTERNATIONAL JOURNAL OF EDUCATION SCIENCE ISSN 2822-6844 Year (2025) Vol:5 Issue: 2 Issued in December 2025 www.e-edusci.com 674 achievement gaps stemming from socioeconomic differences by providing content tailored to students' individual pace, knowledge level, and learning needs. Another significant opportunity is the reduction of cognitive load. AI systems free up students' mental energy from unnecessary tasks, allowing them to focus more on conceptual understanding and higher-order thinking. Furthermore, teacher productivity is significantly increased by AI technologies, as labor-intensive processes such as creating lesson plans, writing questions, conducting assessments, and classifying student errors can be automated, freeing teachers to dedicate more time to pedagogical design. Finally, AI-generated explanations and alternative solutions offer a powerful contribution to the development of mathematical thinking by allowing students to compare and contrast various ways of thinking. These diverse opportunities clearly demonstrate the transformative potential of AI in mathematics education. 5. Limitations and Criticisms Despite the significant opportunities AI offers in mathematics education, there are several limitations and criticisms that must be considered. First, epistemological issues can sometimes cause AI-generated mathematical explanations to be superficial, inaccurate, or inconsistent. The phenomenon of "hallucination," particularly seen in large language models, poses a serious pedagogical risk because misleading students can impair conceptual learning. Furthermore, ethical and data privacy issues are also a significant area of debate. The sensitive nature of student data raises the possibility of improper processing of this information or the introduction of bias through biased algorithms (Williamson & Eynon, 2020). Furthermore, there are concerns that excessive use of AI tools could undermine students' own problem-solving skills and create learning addiction. Finally, the role of the teacher in AI-enhanced learning environments must be redefined because the models for teacher-AI interaction, the distribution of responsibility, and the boundaries of pedagogical control are not yet clear. These limitations necessitate a careful, critical, and robust approach to the use of AI in education. 6. Method Proposal The rapid development of AI applications in mathematics education increases the need for a comprehensive research framework to ensure the pedagogically safe, effective, and sustainable use of these technologies. Accordingly, the first component of the proposed model, pedagogical fit analysis, aims to assess the extent to which AI tools align with mathematics curricula, course objectives, and learning outcomes. This ensures that the use of AI advances without compromising pedagogical foundations. The second component, cognitive process modeling, examines how students' fundamental mathematical cognitive processes, such as conceptual understanding, reasoning, representation transformation, and problem solving, are supported by AI systems. This analysis reveals the ways in which AI contributes to students' cognitive architecture. The third element of the proposed framework, generalizability testing, aims to determine whether the developed AI models demonstrate consistent performance across diverse student profiles, grade levels, and diverse mathematical content. However, AI accuracy validation is also a critical step, as the reliability of solutions, explanations, and problem-solving steps generated by AI systems must be confirmed by independent methods. Another component of the framework, long-term impact measurement, aims to assess how AI use impacts students' learning habits, mathematical thinking skills, and cognitive development processes over time. Finally, the ethical and professional standards framework requires the establishment of fundamental principles such as student data protection, algorithmic fairness, teacher ASES EDUSCI ASES INTERNATIONAL JOURNAL OF EDUCATION SCIENCE ISSN 2822-6844 Year (2025) Vol:5 Issue: 2 Issued in December 2025 www.e-edusci.com 675 responsibilities, and classroom usage boundaries. This framework is essential for the safe, transparent, fair, and pedagogically meaningful use of AI systems in the classroom. It also provides a systematic basis for researchers, teachers, and policymakers to evaluate future AI applications. 7. Conclusion Artificial intelligence offers significant transformation potential in mathematics education. When supported by an appropriate pedagogical framework, ethical principles, and algorithmic verification processes, AI can be a tool that strengthens mathematical thinking, personalizes learning, and enriches the professional roles of teachers. However, uncontrolled use or use devoid of a critical foundation may threaten learning quality and cognitive development. Therefore, the integration of AI into mathematics education requires an interdisciplinary approach, rigorous verification, and pedagogical research. In this context, for the sustainable and effective use of AI in mathematics education, technology developers, educational researchers, teachers, and policymakers must work collaboratively. In particular, preserving a culture of mathematical reasoning and proof, ensuring transparency in algorithmic processes, and continuosusly evaluating classroom applications of AI tools are among the fundamental components of a healthy integration process. Furthermore, increasing teachers' AI literacy and adopting digital tools aligned with pedagogical design principles are critical steps to ensuring that technology adds value to learning processes. Future research should more comprehensively examine the long-term cognitive impact of AIbased systems on mathematical concept learning and how they shape student autonomy and mathematical thinking processes. Furthermore, standardizing explainability, data ethics, security, and model validation mechanisms in educational contexts will contribute to the development of AI applications within a trustworthy, fair, and pedagogically supportable framework. Thus, AI can become not only an innovative tool in mathematics education but also a component that enriches the epistemological foundations of the discipline and expands learning opportunities. ASES EDUSCI ASES INTERNATIONAL JOURNAL OF EDUCATION SCIENCE ISSN 2822-6844 Year (2025) Vol:5 Issue: 2 Issued in December 2025 www.e-edusci.com 676 8. References 1. Drori, I., Zhang, S., Shuttleworth, R., Tang, L., Lu, A., Ke, E., ... & Udell, M. (2022). A neural network solves, explains, and generates university math problems by program synthesis and few-shot learning at human level. Proceedings of the National Academy of Sciences, 119(32), e2123433119. https://doi.org/10.1073/pnas.2123433119 2. Holmes, W., Bialik, M., & Fadel, C. (2019). Artificial intelligence in education: Psromises and implications for teaching and learning. Center for Curriculum Redesign. https://curriculumredesign.org/our-work/artificial-intelligence-in-education/ 3. Kasneci, E., Sessler, K., Eling, N., & Kasneci, G. (2023). ChatGPT for good? On opportunities and challenges of large language models for education. Learning and Individual Differences, 103, 102274. https://doi.org/10.1016/j.lindif.2023.102274 4. Koedinger, K. R., & Aleven, V. (2016). Different futures of adaptive collaborative learning support. International Journal of Artificial Intelligence in Education, 26(1), 1– 14. https://doi.org/10.1007/s40593-015-0083-0 5. Liu, R., & Koedinger, K. R. (2017). Closing the loop: Automated data-driven cognitive model discoveries lead to improved instruction and learning gains. Journal of Educational Data Mining, 9(1), 25–41. https://jedm.educationaldatamining.org/index.php/JEDM/article/view/JEDM912 6. Piech, C., Bassen, J., Huang, J., Ganguli, S., Sahami, M., Guibas, L. J., & SohlDickstein, J. (2015). Deep knowledge tracing. Advances in Neural Information Processing Systems, 28, 505–513. https://papers.nips.cc/paper/2015/hash/5654-deepknowledge-tracing-Abstract.html 7. VanLehn, K. (2006). The behavior of tutoring systems. International Journal of Artificial Intelligence in Education, 16(3), 227–265. https://dl.acm.org/doi/10.5555/1435386.1435388 8. Williamson, B., & Eynon, R. (2020). Historical threads, missing links, and future directions in AI in education. Learning, Media and Technology, 45(3), 223–235. https://doi.org/10.1080/17439884.2020.1798995