A Comprehensive Review of AI-based Intelligent Tutoring Systems: Applications and Challenges

AI-based Intelligent Tutoring Systems (ITS) have significant potential to transform teaching and learning. As efforts continue to design, develop, and integrate ITS into educational contexts, mixed results about their effectiveness have emerged. This paper provides a comprehensive review to understand how ITS operate in real educational settings and to identify the associated challenges in their application and evaluation. We use a systematic literature review method to analyze numerous qualified studies published from 2010 to 2025, examining domains such as pedagogical strategies, NLP, adaptive learning, student modeling, and domain-specific applications of ITS. The results reveal a complex landscape regarding the effectiveness of ITS, highlighting both advancements and persistent challenges. The study also identifies a need for greater scientific rigor in experimental design and data analysis. Based on these findings, suggestions for future research and practical implications are proposed.

Personalized Learning: ITS represent an advanced approach to personalized learning by leveraging sophisticated algorithms to tailor educational experiences to individual learners. Effective personalization [136] in education requires a comprehensive understanding of the learner and the design of meaningful and relevant learning tasks. Given learners’ diverse needs and characteristics, creating an environment that adapts to these variations is inherently complex. The learner model, a critical component of adaptive environments, encapsulates the learner’s profile and serves as the foundation for personalization. By accurately modeling the learner’s needs and characteristics, ITS can adjust the learning environment dynamically, resulting in a more effective and efficient educational experience.

In AI-based ITS, personalized learning involves customizing educational content and pacing to match each student’s unique abilities and requirements. This targeted approach addresses individual learning gaps and leverages student strengths, leading to more effective knowledge acquisition and retention. For example, specific algorithms like Bayesian networks [53], decision trees [52], and deep learning models are employed to continuously update and refine the learner model [51], allowing for real-time adaptation of instructional strategies. Empirical studies consistently show that students in personalized learning environments outperform their peers in standardized settings [17], mainly due to the tailored support they receive, which aligns with their specific learning needs.

Kochmar et al. [207] demonstrated this with the Korbit learning platform, a largescale, dialogue-based ITS. This model meets personalization by generating automated, data-driven personalized feedback using advanced ML and NLP techniques. The system dynamically adapts its instructional content based on the student’s individual needs, providing personalized hints, explanations, and mathematical guidance. The system can tailor educational interventions according to the learner’s profile and their zone of proximal development (ZPD), optimizing learning outcomes and enhancing student engagement by employing models such as neural networks and reinforcement learning algorithms.

Bernacki et al. [189] systematically explored and clarified the multifaceted landscape of personalized learning (PL) by examining who is responsible for personalizing learning, what elements are personalized, how personalization is implemented, and the purposes it serves. The article underscores the variability in definitions and implementations of PL across different educational contexts, complicating the assessment of how effectively PL designs leverage student characteristics to enhance learning outcomes. Laak et al. [101] implemented a dynamic student modeling approach combined with the zone of proximal development (ZPD) concept. In their ITS, the student model is continuously updated based on the learner’s interactions, capturing cognitive states, knowledge levels, and affective states. By aligning instructional content within the learner’s ZPD, the ITS ensures that tasks are both challenging and achievable, optimizing learning outcomes and fostering deeper engagement. This approach integrates insights from cognitive science and educational psychology, creating a more effective and personalized learning environment that directly addresses the limitations of traditional, one-size-fits-all tutoring methods.

Shah et al. [205] and [10] explored the use of LLMs to enhance personalization within ITS by enabling more natural and dynamic interactions between the system and learners. LLMs, such as those in the GPT series, are trained on vast datasets and are incorporated into ITS to understand and generate human-like responses. This capability allows ITS to provide individualized feedback, clarify concepts, answer questions, and engage in dialog-based tutoring sessions that adapt to the learner’s needs. By leveraging the advanced capabilities of LLMs, these systems overcome traditional tutoring’s limitations, such as scalability and consistency, offering a more flexible and responsive learning experience.

Adaptive Learning : Systems that include ITS [110] [112] and Adaptive ITS (AITS), dynamically adjust learning paths, pacing, and content to meet each student’s unique needs and preferences. By integrating adaptive features, these systems ensure that learning experiences are efficient and personalized. Adaptive mechanisms, such as realtime assessments of student progress, are employed to adjust instructional content and difficulty levels, maintaining an optimal challenge that prevents boredom from overly simple material and frustration from challenging content [150]. This adaptability keeps students engaged and motivated, especially in diverse environments where students have varying backgrounds and levels of prior knowledge. Furthermore, the long-term effects on student motivation and self-regulated learning are significant, as these systems foster sustained interest and autonomy by tailoring materials to each student’s progression level [135] [124].

Phobun et al. [98] explained that AITS seamlessly integrate ITS with Adaptive Hypermedia (AH) to create personalized and dynamic learning experiences. AITS are composed of an expert model that adapts educational content and delivery based on a learner’s current knowledge, learning style, and progress. By leveraging adaptive presentation and navigation, AITS ensure that each learner receives the most relevant material optimally. This real-time adaptability, grounded in a deep understanding of learner interactions, enhances theoretical and practical learning, making AITS a powerful tool for personalized education across various domains.

To personalize learning, Liu et al. [187] employed adaptive prompts in ITS grounded in educational theories. These prompts decompose complex problems, foster critical thinking, and offer tailored cognitive support. By dynamically adjusting to student performance, ITS enhance problem-solving efficiency and deepen understanding, ensuring a highly personalized and practical learning experience.

Learner Modeling: Learner models play a critical role in ITS by representing essential user-specific information. As discussed by [110], user models can include the overlay approach, where user knowledge is mapped as a subset of expert domain knowledge, and the uncertainty-based approach, which uses probabilistic methods to manage the uncertainties in understanding user behaviors and preferences. This enables ITS to finely tune the content, navigation, and interface to each user’s needs. Paredes et al. [112] introduce a theoretical framework for dynamically modeling student learning styles, combining explicit data from questionnaires and implicit data from ongoing course interactions. Learner modeling involves creating detailed representations of student’s cognitive and emotional states. These models enable ITS to provide targeted interventions, such as adjusting the complexity of tasks or offering additional emotional support when frustration is detected. By accurately identifying these challenges, ITS can help students overcome specific obstacles, leading to a more supportive and effective learning experience. Enhanced learner modeling can also include emotional [90] and cognitive state recognition [89] to provide real-time, context-sensitive responses, thereby improving student engagement and well-being. Kumar et al. [147] argue that accurately modeling and understanding the learner’s characteristics and behaviors allows ITS to deliver a highly personalized and effective educational experience. This adaptivity is achieved through continuous assessment and real-time adjustments based on the learner’s interactions and progress.

Recommender Systems: The application of recommender systems within Technology Enhanced Learning (TEL) is crucial, as explored by [113]. These systems help find relevant learning resources, enhancing the learning experience by providing personalized content recommendations. Recommender systems in ITS suggest personalized learning resources, enhancing educational inclusivity by catering to diverse learning styles and cultural backgrounds. However, there is a risk of reinforcing biases if the recommendation algorithms are not carefully designed. To address this issue, it is crucial to ensure that the data used for training these systems is both diverse and representative. Additionally, regular audits and adjustments of the recommendation criteria help maintain fairness and inclusivity. Ensuring diversity and inclusivity in these recommendations can prevent content bias and support a broader educational scope. Muangprathub et al.[123] focused on developing a learning recommendation component within ITS that can dynamically predict and adapt to individual learners’ styles. The primary focus is on creating an adaptive algorithm and an improved knowledge base that enables personalized learning by providing tailored content recommendations.

Real-Time Feedback and Assessment: One of the most important advantages of AI-based ITS is the ability to provide immediate feedback [156] [20]. These systems assess student responses in real-time and offer corrective feedback [159], allowing students to learn from their mistakes quickly. Effective student feedback is a critical tool for improving the quality of education by enhancing instructional methods, refining curricula, and ultimately improving learning experiences and outcomes [61]. The impact of feedback depends on how it is processed and utilized by institutions. Positive feedback can boost student confidence, while negative feedback, if not delivered thoughtfully, can lead to anxiety or decreased motivation. To minimize these adverse effects, ITS should employ strategies such as constructive criticism [50], emphasizing effort over inherent ability, and providing balanced feedback highlighting strengths and improvement areas. Careful design is essential to avoid reinforcing negative self-perceptions or anxiety, particularly in formative assessments. This strategic use of feedback ensures continuous improvement in education, making the learning process more responsive to student needs and more effective overall [137]. LLMs enhance feedback in ITS by providing personalized, real-time responses that adapt to individual learner needs. These models can offer nuanced and empathetic feedback, improving the learning experience. However, for LLMs to be truly effective, their use must be grounded in solid theoretical frameworks and validated through empirical research to ensure that their feedback effectively supports student learning.

Data-Driven Insights and Explainable AI (XAI): The integration of data-driven insights [184] in ITS significantly enhances the transparency and explainability of the system’s decision-making processes. ITS can clarify these processes by employing mechanisms such as XAI, thereby building user trust. Data-driven insights contribute to transparency by offering clear, evidence-based explanations for instructional decisions. Implementing XAI in ITS involves using models that provide understandable rationales for their recommendations and actions. Best practices include offering users accessible explanations of how decisions are made and ensuring that these systems are interpretable by educators, students, and parents, thus fostering trust and accountability.