The Future of Education: A Systematic Literature Review of Self-Directed Learning with AI

Abstract

As digital transformation continues to redefine education, understanding how emerging technologies can enhance self-directed learning (SDL) becomes essential for learners, educators, instructional designers, and policymakers, as this approach supports personalized learning, strengthens student autonomy, and responds to the demands of more flexible and dynamic educational environments. This systematic review examines how artificial intelligence (AI) tools enhance SDL by offering personalized, adaptive, and real-time support for learners in online environments. Following the PRISMA 2020 methodology, a literature search was conducted to identify relevant studies published between 2020 and 2025. After applying inclusion, exclusion, and quality criteria, 77 studies were selected for in-depth analysis. The findings indicate that AI-powered tools such as intelligent tutoring systems, chatbots, conversational agents, and natural language processing applications promote learner autonomy, enable self-regulation, provide real-time feedback, and support individualized learning paths. However, several challenges persist, including overreliance on technology, cognitive overload, and diminished human interaction. These insights suggest that, while AI plays a transformative role in the evolution of education, its integration must be guided by thoughtful pedagogical design, ethical considerations, and a learner-centered approach to fully support the future of education through the internet.

1. Introduction

The accelerated growth of the Internet of Things (IoT) has opened new opportunities for the future of education by enabling seamless connectivity among devices, services, and data sources. This digital network has expanded the dissemination of knowledge and improved access to educational resources across diverse geographic contexts. As a result, more flexible, interactive, and inclusive online learning environments are being promoted, although their effective implementation still depends on factors such as technological infrastructure, teacher preparedness, and equitable access [1].

In this context, the Education 4.0 model and the emerging Education 5.0 approach are driving a profound transformation in how teaching and learning are conceptualized. Both advocate for a pedagogy focused on personalization, adaptability, and collaboration between humans and intelligent systems. Education 5.0, in particular, envisions a future where the interaction between human intelligence and artificial intelligence enhances students’ cognitive capacities, going beyond the boundaries of traditional approaches [2].

As education continues to shift to digital environments, it becomes essential for students to develop autonomy in managing their learning processes. Among the most effective methodologies to foster this ability is self-regulated learning (SRL), a key strategy in both face-to-face and virtual education that promotes independence and self-management [3]. The rise of online learning has increased students’ control over their learning paths, which has, in turn, heightened the need to cultivate robust self-directed learning (SDL) skills [4].

Artificial intelligence (AI) has garnered growing interest in educational research due to its capacity to simulate human cognitive processes such as reasoning, planning, and decision-making. Its application in education has led to the development of innovative tools such as intelligent tutors, data mining systems, natural language processing technologies, and educational robots [5]. These technologies are being implemented across various levels and disciplines, enabling educators to respond more precisely to individual learner needs. AI can analyze data on students’ interests, performance, and learning behavior to tailor content and activities to their individual profiles [6].

In addition to facilitating personalization, AI automates tasks such as exam grading and instant feedback, thereby contributing to a more dynamic and efficient learning experience [7]. This supports learner autonomy and encourages active participation in digital environments, which are fundamental pillars of self-directed learning.

A positive attitude toward the use of digital technologies largely depends on the perceived usefulness of these tools and users’ digital competence. When individuals are familiar with technological tools and perceive them as beneficial, adoption becomes more active and sustainable, facilitating the integration of these technologies into learning and other domains [8].

Background

The formal theory of self-directed learning originates in the field of adult education, and one of the first theorists to address it systematically was Malcolm Knowles [9], who in his 1975 book Self-Directed Learning: A Guide for Learners and Teachers laid the conceptual foundations of this approach. Knowles argues that self-directed learning is a process in which individuals take the initiative either independently or with the assistance of others to diagnose their learning needs, set goals, identify resources, choose and implement appropriate strategies, and evaluate their outcomes. In his work, he emphasizes that this type of learning is particularly relevant in adulthood, when individuals seek greater autonomy, flexibility, and responsibility for their own development, representing a fundamental shift from traditional teacher-centered educational models [10].

Another scholar who contributed in parallel to Knowles’s work was Allen Tough, who also provided one of the earliest comprehensive descriptions of self-directed learning [11]. Tough concluded that adults devote a significant amount of time to what he termed “learning projects,” aimed at acquiring, maintaining, or modifying personal skills, knowledge, or attributes [12]. These projects do not necessarily occur within formal educational environments but can take place through various activities such as reading, listening, observing, attending courses, reflecting, or engaging in deliberate practice. His work offered a broader understanding of how adults manage their own learning processes.

SDL, as defined by Kim et al. [4], is a process in which individuals take control and responsibility for their own learning, guided by a structure that enables them to achieve educational goals. According to different authors, it can be understood as a personal trait, an active process, or a condition influenced by the learning environment [11,12,13]. It is regarded as a key pathway to lifelong learning, as it enables individuals to identify their learning needs, set objectives, seek resources, apply strategies, and evaluate outcomes [3].

The SDL process begins with the formulation of clear goals, followed by the search for appropriate resources—such as articles, books, and online courses—and the execution of activities aligned with the intended objectives. Throughout this process, continuous monitoring and assessment help detect progress and allow for strategic adjustments, ensuring ongoing improvement [4].

Implementing SDL successfully in digital environments presents several challenges, particularly due to the lack of direct support, low motivation, or limited use of technological resources [14]. In this context, AI can serve as a significant support tool, offering asynchronous assistance and enhancing constructivist practices such as inquiry, collaborative work, and reflection [15].

Analyzing how different studies address the relationship between online self-regulated learning and artificial intelligence (AI)-based tools reveals a growing interest in integrating emerging technologies that promote autonomy and personalization in digital educational environments. These studies demonstrate that the use of intelligent systems can significantly strengthen self-directed learning (SDL).

For example, the study by Adam et al. [16] examined the application of SRL in virtual education settings, particularly in mathematics, through a systematic review of 130 articles published between 1986 and 2017. The results show that SRL, understood as a cyclical process of planning, execution, and self-reflection, is essential for improving online academic performance. Strategies such as metacognition, resource management, self-assessment, and self-motivation enable students to effectively regulate their learning, and their application in virtual environments is associated with higher engagement, autonomy, and achievement.

Gambo and Zeeshan [17] conducted a systematic review of 15 articles published between 2012 and 2020 to investigate how virtual learning environments (VLEs) support the development of SRL. The results indicate that VLEs promote SRL primarily through features such as automated feedback, progress tracking, and personalized instruction. These tools assist students in planning, monitoring, and evaluating their learning, thereby strengthening their autonomy.

Similarly, the study by Viberg et al. [18] explored how AI-based technologies support SRL in higher education virtual environments, through a systematic review of 54 publications from 2011 to 2019. The findings suggest that most AI tools focus on automated feedback, personalized content, and student progress monitoring. These technologies contribute to the development of SRL skills such as planning, self-evaluation, and motivation control, although the study highlights the need for more pedagogically grounded and student-centered designs.

Regarding the specific use of AI tools in this type of learning, Alshahrani [19] examined AI-based educational chatbots and found that they can provide personalized support, improve comprehension, increase motivation, and foster student engagement. They also optimize teachers’ time and enable more flexible instruction, especially in high-demand contexts such as higher education. However, the study recommends further research to establish guidelines that maximize their benefits and address the ethical and practical challenges of their use.

Ma et al. [20] conducted a review of 37 studies to explore who uses AI, for what purposes, and how it is implemented in teaching. They found that university students are the primary users and employ these tools for academic support, writing assistance, and oral practice. While the evidence indicates benefits in terms of personalization and motivation, it also underscores the need for well-designed methodologies to achieve significant impacts on self-directed learning.

Likewise, Younas et al. [21] evaluated AI technologies aimed at supporting SDL and their influence on educational outcomes. Their findings reveal that systems such as intelligent tutors and conversational agents provide real-time feedback and foster autonomy. However, they also identified risks related to technological dependency and cognitive overload, highlighting the need for further research to optimize their use.

As evidenced in the literature, various studies agree that the intersection between artificial intelligence (AI) and machine learning is shaping a new era of education that is more autonomous, adaptive, and student-centered. Understanding this relationship is essential for designing learning ecosystems that are more resilient, inclusive, and sustainable in the long term. However, despite the exponential growth in the development of AI-based tools over the past decade, academic literature shows considerable fragmentation in how these technologies are classified, applied, and evaluated within educational contexts.

This dispersion makes it difficult to identify trends, compare approaches, and support evidence-based decision-making by educators, educational technology developers, and institutions. Although some studies focus on specific applications, such as the works of Alshahrani [19], Younas et al. [21], and Lin [22], which explore the use of chatbots and conversational agents like ChatGPT, the absence of a systematic analytical framework limits the ability to comprehensively understand the field and its most significant developments.

Moreover, while self-directed learning (SDL) is recognized as a key competence in contemporary virtual environments, its connection with artificial intelligence technologies has not always been addressed explicitly or in sufficient depth. In this regard, the Key Competences for Lifelong Learning [23] proposed by European Commission Directorate-General for Education, Youth, Sport and Culture state that individuals must be able to access, acquire, process, and assimilate new knowledge and skills throughout their lives, which requires effective management of their own learning processes. This ability demands self-discipline, organization, and a willingness to dedicate time to autonomous study. Such competences reflect the need to develop active, responsible, and adaptable learners who can self-regulate their learning in increasingly dynamic and technology-mediated educational contexts.

In this context, there is a clear need for a systematic review that integrates and critically analyzes the available evidence. This study aims to identify which technologies are being used to enhance SDL, while also exploring why these tools are relevant, how they are embedded in different stages of the learning process, and what pedagogical, ethical, and practical implications they entail. The review provides an updated overview of the state of the art and seeks to guide future research and implementation efforts for learners, educators, instructional designers, and policymakers, with the goal of promoting AI-based educational strategies that support autonomy, self-regulation, and equity in modern education.

To this end, the study poses the following research questions:

• What artificial intelligence tools or applications can be used in self-directed learning?
• What are the main contexts or goals in which AI is used for self-directed learning?
• In which stages or processes of self-directed learning is AI integrated?
• What are the main issues or challenges associated with the use of AI in self-directed learning?
• What are the potential implications of using artificial intelligence in the future development of online education?

2. Materials and Methods

This research included a systematic literature review based on the PRISMA guidelines (Preferred Reporting Items for Systematic Reviews and Meta-Analyses), an international standard that promotes transparency, methodological rigor, and reproducibility in scientific study reviews [24]. The PRISMA methodology structures the review process into four main phases: identification, screening, eligibility, and inclusion, enabling the objective and well-founded selection of relevant studies on the impact of artificial intelligence in education.

It is worth highlighting that all procedures, data, and pertinent information related to the project have been preserved and are openly accessible via the Open Science Framework (OSF) Registries. Details on how to access these materials are provided in the document’s “Data Availability Statement” section. This ensures transparency and open access to the project’s resources.

2.1. Eligibility Criteria and Information Sources

The information sources selected for this review were the Scopus and Web of Science databases, both internationally recognized for their scientific rigor, indexing standards, and broad interdisciplinary coverage. These databases were chosen due to their inclusion of high-impact journals, peer-reviewed literature, and comprehensive metadata that facilitate advanced bibliographic analysis.

The inclusion criteria comprised publications from the previous ten years (2015–2025), with the aim of ensuring the relevance and currency of knowledge in a dynamic field such as artificial intelligence applied to education. Additionally, studies published in any language were considered in order to provide a global perspective, and documents such as scientific articles, book chapters, conference proceedings, and review papers were included. The exclusion criteria required the removal of errata, corrections, and duplicate records, as well as the exclusion of emerging low-impact journals from the Web of Science database, prioritizing well-established and high-quality publications.

2.2. Search Strategy

The search strategy was designed in accordance with the PRISMA 2020 guidelines. Initially, a preliminary search was conducted in the Scopus database using a combination of controlled and uncontrolled terms related to “artificial intelligence” and “self-regulated learning.” Based on the identified records, an analysis of indexing terms and keywords was performed using R Studio version 4.4.2, which enabled the refinement and expansion of search expressions. This optimization focused on capturing relevant terminological variations and enhancing search sensitivity.

Subsequently, the strategies were specifically adapted to the syntax, Boolean operators, document filters, and search fields of each database (Scopus and Web of Science). The filtering parameters included restriction by document type (articles, book chapters, conference proceedings, and reviews), the publication year range (2020–2025), and the exclusion of emerging low-impact sources in Web of Science (see

Table 1. Search Strings and Results Obtained.

Database	Search Strings	Total
Scopus	(TITLE (“artificial intelligence” OR “AI” OR “brain-like intelligence” OR “Machine learning” OR “deep learning” OR “learning algorithms” OR “chatgpt” OR “chatbot” OR “neural networks”) AND TITLE-ABS-KEY (“self-directed learning”)) AND PUBYEAR > 2019 AND PUBYEAR < 2026 AND (LIMIT-TO (DOCTYPE, “ar”) OR LIMIT-TO (DOCTYPE, “cp”) OR LIMIT-TO (DOCTYPE, “ch”) OR LIMIT-TO (DOCTYPE, “re”))	110
Web of Science	((TI = (“artificial intelligence” OR “AI” OR “brain-like intelligence” OR “Machine learning” OR “deep learning” OR “learning algorithms” OR “chatgpt” OR “chatbot” OR “neural networks”)) AND TS = (“self-directed learning”))	42
		153

).

2.3. Study Selection Process

Initially, the records retrieved from the Scopus and Web of Science databases were combined, resulting in a total of 153 entries. Duplicate records were subsequently removed, reducing the number of unique documents to 114. In the second phase, the titles, abstracts, and keywords of each record were reviewed, leading to the exclusion of six studies that were not directly related to the research topic. An additional six documents were excluded due to lack of access to the full text.

The remaining studies underwent a methodological risk of bias assessment, using criteria adapted to the type of research, in order to ensure the inclusion of studies with a low risk of bias. It is important to note that the entire process of screening, eligibility assessment, and decision-making was conducted collaboratively by all members of the research team. Discrepancies were resolved through consensus-based discussion, ensuring transparency, objectivity, and methodological rigor in the study selection process.

Ultimately, 77 primary studies that met the quality, thematic relevance, and methodological rigor criteria were selected, forming the final corpus analyzed in this systematic review. The complete process is summarized in Figure 1, presented in accordance with the PRISMA flow diagram.

2.4. Bias Risk Assessment

The risk of bias was assessed by adapting the criteria established by the Cochrane Collaboration. Aspects such as random sequence generation, allocation concealment, blinding of participants and assessors, data integrity, and the potential for selective outcome reporting were evaluated. This analysis was adjusted according to the type of study reviewed, acknowledging that not all criteria are uniformly applicable across different methodologies. This assessment enabled the selection of studies with a low risk of bias, thereby strengthening the reliability of the review

2.5. Synthesis Methods

Finally, for data synthesis, relevant information from each study was extracted in relation to the formulated research questions. As some answers were not explicitly stated and certain ideas were presented in varied ways, a thematic classification process was carried out. Responses were organized into conceptual categories to clarify and simplify the presentation of results. The detailed methodology for the coding and categorization applied during the synthesis is described in the Supplementary Materials for reference.

3. Results

3.1. Bibliometric Data

Figure 2 presents the top ten countries with the highest number of documents included in the systematic review on artificial intelligence tools for self-directed learning. China (59), the United States (47), India (23), and South Korea (20) stand out as the main contributors, followed by Germany (11), Greece (8), Indonesia and Thailand (7 each), and Malaysia and Turkey (6 each).

Asian countries—particularly China, India, and South Korea—dominate the research output in this field, followed by traditional powerhouses such as the United States and Germany. This pattern may reflect national innovation policies as well as a growing interest in personalized learning supported by AI technologies. Although these countries had the highest representation, contributions from other nations were also identified, albeit to a lesser extent, indicating a growing global interest in the use of artificial intelligence to support self-directed learning.

The co-occurrence network (Figure 3) visually shows the key terms that appear most frequently in the scientific literature on artificial intelligence applied to self-directed learning. At the center of the map, the terms “artificial intelligence”, “self-directed learning”, and “students” stand out, indicating that they are the most addressed and interconnected concepts. Around these core terms, three thematic communities can be observed.

In Figure 3, red dots refer to technical and pedagogical terms such as machine learning, deep learning, engineering education, and chatbots, which reflect approaches from computer and educational sciences; blue dots indicate concepts related to human aspects such as human, humans, medical education, nursing, and questionnaires, suggesting studies focused on clinical or social contexts; green is used for emerging terms like federated learning, self-supervised learning, and contrastive learning, which represent advanced technological trends. This structure evidences the multidisciplinarity of the field, with a strong link between emerging technologies, educational methodologies, and human aspects, highlighting the centrality of AI in the process of autonomous learning.

3.2. Artificial Intelligence Tools or Applications Used in Self-Directed Learning

Table 2. Artificial intelligence tools or applications used in self-directed learning.

Category of AI Tools	Especific AI Tools	Studies That Mention It
General Conversational Assistants	ChatGPT	[3,8,13,14,19,22,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46]
	Bing Chat, Ernie Bot, CoolE Bot, AI Conversational Chatbot (AICC), Dialogflow Chatbot (DC), SIA, Google Assistant, Voicebots	[5,21,29,44,47,48,49]
AI Educational Assistants	TeacherGAIA, AGILEST Agents, Duolingo (with AI), EduMentor, Khanmigo (Khan Academy)	[15,35,50,51]
Messaging Applications with Integrated AI	Line	[52]
Smart Devices with Voice Assistants	Alexa, Homepod, Google Nest	[53]
Intelligent Tutoring Systems (ITS) and Related Models	Intelligent Tutoring Systems (ITS), Fuzzy-Based ITS (FB-ITS), Meta Tutor (MT-ITS), BioWorld—ITS, Open Learner Model (OLM), EnSmart (LMS with ITS), RADIAL (Radiology’s Intelligent Adaptive Learning), National Imaging Informatics Course—Radiology (NIIC-RAD)	[21,54,55,56]
Creative or Visual Assistance Tools with AI	AutoDraw, AMCAD Artificial intelligence-enabled evaluating for computer-aided drawings, Gamma (visual assistant for presentations), Waste Genie App (environmental/visual education with AI)	[57,58,59,60]
Writing and Language Support Tools	Quillbot, DeepL Write, LanguageTool	[25,45]
AI-Assisted Evaluation or Diagnostic Platforms	EXAIT (Educational eXplainable AI Tool), AI-Scorer, RADHawk (RH), AMCAD	[6,54,57,61]

presents the artificial intelligence tools or applications used in self-directed learning, according to the studies analyzed in the review. In order to facilitate their understanding and analysis, these tools were classified into functional categories, considering their main purpose and the type of interaction they offer to the user. This classification enabled a clearer visualization of the diversity of existing applications and how they relate to different educational needs, without intending to establish usage patterns or specific areas of application.

The most prominent category is that of general conversational assistants, with ChatGPT being the most widely used tool. Its predominance is due to its high accessibility, versatility, and ability to generate coherent responses, detailed explanations, and personalized feedback on a wide range of topics, making it a valuable resource for self-directed learning. Other assistants follow, such as Bing Chat, Ernie Bot, CoolE Bot, Dialogflow, Google Assistant, and SIA, which also enable interaction through natural language, although with more limited or specific approaches.

In second place are AI educational assistants, specifically designed for teaching and learning contexts. Tools such as Khanmigo (from Khan Academy), Duolingo with AI, EduMentor, and TeacherGAIA provide content tailored to the student’s level, progress tracking, and personalized suggestions, fostering autonomous and student-centered learning.

The category of messaging applications with integrated AI, represented by LINE, shows a more specific use, allowing the integration of educational assistance functions within an everyday communication platform.

Smart devices with voice assistants such as Alexa, Homepod, and Google Nest provide indirect support for self-directed learning by facilitating quick searches, reminders, and access to educational content through voice commands, although their focus is not strictly educational.

In the category of intelligent tutoring systems (ITSs) and related models, tools such as Meta Tutor, BioWorld, Open Learner Model, and RADIAL provide adaptive teaching based on continuous analysis of student performance. These systems are highly personalized and are frequently used in specialized contexts such as medicine or applied sciences.

Creative or visual assistance tools with AI, such as AutoDraw, Gamma, and AMCAD, support the development of graphic, design, or visual interpretation skills, while the Waste Genie app focuses on environmental education. These tools allow students to interact with visual content autonomously, which is useful in technical or creative disciplines.

Regarding writing and language support tools, Quillbot, DeepL Write, and LanguageTool stand out, offering grammar correction, translation, and rephrasing functions, facilitating text production in self-learning processes, especially in foreign languages or academic writing.

Finally, AI-assisted evaluation or diagnostic platforms such as EXAIT, AI-Scorer, and RADHawk are designed to analyze student performance, generate automatic grades, or perform diagnostics in specific areas such as radiology or technical design. Their use is valuable in contexts where quick and objective feedback is required.

3.3. Main Contexts in Which AI Is Used for Self-Directed Learning

Figure 4 presents the main contexts in which artificial intelligence is used for self-directed learning, covering various areas, educational levels, and training processes. These include higher education and health training, language learning, secondary and primary education, as well as continuing education, technical training, and virtual learning, reflecting the breadth of scenarios in which AI supports the autonomous development of knowledge.

The predominant context is higher education, with a significant difference compared with the other contexts, indicating that most research has focused on universities and tertiary-level institutions. This can be attributed to the fact that students at this level often have greater autonomy and access to advanced technologies, which facilitates the implementation of AI in self-directed learning.

Health training, specifically in medicine and nursing, appears as one of the most relevant contexts for the application of AI in self-directed learning, ranking second after higher education. This is due to the high demand for highly personalized learning in these fields, based on clinical practice, decision-making, and constant knowledge updates. Artificial intelligence supports this process through simulations, immediate feedback, intelligent tutoring systems, and diagnostic evaluation [32,33,34,62,63], allowing students to develop competencies at their own pace and in safe environments [30]. Moreover, the use of AI in these contexts supports continuous training, repetition of complex clinical scenarios, and improvement of critical skills, which explains its growing adoption in health programs [3,63,64].

Language learning also stands out as one of the main contexts for the application of artificial intelligence in self-directed learning, reflecting significant interest in the studies analyzed. This is because AI allows for personalized teaching according to the student’s level, pace, and learning style, providing immediate feedback, voice recognition, grammar correction, and adaptive exercises [13,14,37,39,43,50,65]. Tools such as conversational chatbots, intelligent translators, and gamified platforms enhance autonomous practice in comprehension, oral and written expression, creating an immersive environment even outside the classroom [5,8,29,47,48].

Following this, secondary education and continuing education and professional development also show a growing use of AI, especially in competency update programs and technical training. In continuing education and professional technical training, AI has gained particular relevance in programs aimed at adult learners who require flexibility to study according to their availability [21,22,51,66]. Furthermore, AI use allows for the creation of personalized learning pathways, identification of learning gaps, and improved knowledge retention, thus contributing to continuous professional development and employability in changing work environments.

To a lesser extent, but still present, are contexts in primary education [15,49,67], various educational levels, and virtual and distance learning [19,31,56], suggesting that, although there is interest in these areas, the integration of AI for self-directed learning is still in its early stages or faces greater technical and pedagogical challenges.

3.4. Self-Directed Learning Processes in Which AI Is Integrated

Table 3. Self-Directed Learning Processes in Which AI has been Integrated.

Processes	Specific Activities	Studies That Mention It
Feedback	Personalized feedback, intelligent feedback, real-time adjustments to the learning process	[8,13,20,27,28,36,45,46,47,48,50,51,66,68]
Tutoring and Mentoring	Personalized tutoring, automated guidance and orientation, assistance in clarifying doubts, adaptive motivation, action recommendations	[3,19,26,33,34,39,49,61,63,69,70,71,72,73,74]
Evaluation and Self-Assessment	Performance evaluation, automated task correction, diagnostic and formative assessment, AI-assisted self-assessment	[5,25,26,30,34,39,52,57,61,62,65,75,76,77]
Learning Planning	Goal setting, content organization, activity planning, design of personalized learning plans, task personalization, student profile diagnosis	[8,15,27,30,31,36,58,65,68,71,72,78,79,80,81]
Execution and Practical Application	Practical application of content, development of adaptive exercises, guided autonomous activities	[8,21,29,30,36,43,47,59,64,82]
Research and Exploration	Resource search, guided research, promoting critical thinking, exploration of topics, development of problem-solving skills	[3,21,33,38,40,43,44,49,54,66,67]

presents the main self-directed learning processes in which artificial intelligence has been integrated, along with a brief description of the specific activities associated with each process and some studies that address them. The process in which AI is most applied is feedback, due to its ability to provide immediate, personalized, and continuous responses that guide the student in correcting and improving their performance. Following this, tutoring and mentoring come next, where AI acts as a personalized guide through virtual assistants and intelligent tutoring systems. Relevant applications are also identified in evaluation and self-assessment processes, facilitating the automatic generation of results, as well as in learning planning, through the design of personalized learning pathways. Finally, AI also supports the execution and practical application of knowledge, as well as research and exploration, providing access to sources, recommendations, and real-time information analysis.

3.5. Challenges Associated with the Use of AI in Self-Directed Learning

Some of the main challenges in using artificial intelligence tools for self-directed learning relate to privacy, security, and ethics. There are significant concerns about the protection of students’ personal data, as well as the risks of algorithmic biases that can be cultural or linguistic in nature [39,62,65,68,74,76,77,80,81]. Additionally, there is a lack of clear regulation that addresses the ethical and legal considerations surrounding the use of these technologies [14,26,59,64]. Moreover, the risk of manipulation or misuse of systems, such as jailbreaking or breaches in academic integrity, has been highlighted [15,25,31,39].

Another critical aspect is technological dependence and the potential reduction of student autonomy [27,28,42,43,45,66]. Excessive use of AI can hinder the development of fundamental skills such as critical thinking, independent inquiry, or informed decision-making [32]. It may also reduce meaningful human interaction during the learning process, which is essential for holistic development [76]. The absence of teacher supervision and individual responsibility, coupled with access issues and digital gaps, exacerbates this challenge [34,52].

In terms of learning quality and the accuracy of information, AI-based tools may generate inaccurate, irrelevant, or even misleading content [28,32,35,38,41,46]. The feedback they offer may be limited or standardized, without adapting to the specific needs of each student. A lack of content and response diversity has also been reported, as well as the presence of “hallucinations”—incorrect responses that AI presents with apparent certainty.

From a technical and pedagogical perspective, structural limitations such as connectivity failures, command recognition issues, or general technical problems are evident [5]. Moreover, students may experience cognitive overload or difficulty in personalizing their learning in loosely structured contexts [8,21,56]. Another important obstacle is the lack of training in prompting skills or effective writing of instructions, which limits the full potential of these tools [22,35,40,55]. This underscores the need for rigorous pedagogical design focused on usability and user experience [20,51].

Lastly, challenges related to motivation, engagement, and learning effectiveness have been identified. Some students perceive AI tools as unhelpful, which affects their involvement in the learning process [55,82]. Maintaining long-term motivation is a recurring difficulty, as is the variability in the time students dedicate to self-directed study. Additionally, there is a risk of setting unchallenging goals, which can compromise the quality of the learning achieved [7].

As evidenced in the results, both benefits and challenges are associated with the use of artificial intelligence in self-directed learning. To synthesize and present this information clearly, Figure 5 introduces a SWOT matrix (strengths, weaknesses, opportunities, and threats), which systematically summarizes the key findings of the study. This framework offers a comprehensive overview of the positive contributions of AI—such as learning personalization, real-time feedback, and applicability across diverse educational contexts—while also highlighting its limitations and potential risks, including technological dependency, ethical and legal concerns, and the reduction of meaningful human interaction. The matrix provides a strategic perspective that may inform future research, pedagogical design, and policy development aimed at the responsible and effective integration of AI in self-directed learning environments.

The integration of artificial intelligence (AI) in self-directed learning (SDL) presents several strengths. AI tools are accessible, versatile, and adaptable to various educational contexts. They support personalized learning by adjusting content to students’ level, pace, and style, and offer immediate feedback that enhances self-regulation. A wide range of tools are available for tutoring, assessment, writing, planning, and exploration, with proven effectiveness in higher education, healthcare, language learning, and continuing education.

AI also creates significant opportunities. When combined with active pedagogical models, it can transform learning experiences and promote autonomy, critical thinking, and self-management. It supports personalized professional development by identifying knowledge gaps and aligning learning with labor market needs. Additionally, its use is expanding into secondary, primary, and technical education, with immersive and adaptive environments enhancing engagement and flexibility.

However, some weaknesses must be addressed. Overreliance on AI can reduce critical thinking and learner autonomy. Tools may offer limited or inaccurate feedback, and poorly structured use can lead to cognitive overload. Many learners also lack the skills to effectively interact with AI, such as crafting meaningful prompts.

Several threats also emerge. Ethical and legal concerns include data privacy, bias, and misuse. Unequal access to technology reinforces digital divides. Reduced human interaction may affect socioemotional development, and the absence of clear regulations raises concerns about transparency. Finally, there is a risk of functional dependency, where learners rely excessively on automation and weaken their capacity for independent thinking.

3.6. AI in the Future of Online Education

The progressive integration of artificial intelligence (AI) into self-directed learning processes is shaping a new educational paradigm that will transform, in the near future, how online education is conceived. Tools such as ChatGPT, LanguageTool, or intelligent conversational systems automate tasks like text correction, idea generation, and immediate feedback [25,26,42,76], and are configuring educational environments that are more personalized and responsive to the individual needs of learners [55]. Looking ahead, this capacity to adapt teaching in real time according to students’ performance, interests, and goals promises to establish more effective, flexible, and autonomy-centered learning environments [3,63,72].

One of the most significant future impacts could be the strengthening of learner autonomy. Instead of relying solely on human instructors, students will be able to use intelligent systems to organize their learning goals, access relevant materials, receive strategic feedback, and reflect on their progress [26,27,30,32,56,62,76]. This transformation anticipates a more dynamic and adaptive educational landscape, although it will also require more sophisticated skills in self-management, critical thinking, and responsible technology use.

Artificial intelligence may also guide students more efficiently along their path toward professionalization [3,51,62]. Unlike traditional educational models, which often follow outdated curricula with content of little relevance to current challenges, AI can identify the competencies demanded by the modern labor market and propose personalized learning trajectories [22,52]. This guidance could significantly reduce the time spent on low-value content and instead foster learning that is more applicable to real life, focused on concrete and updated skills.

However, the widespread deployment of AI presents challenges that must be considered from now on. Excessive delegation of decision-making to algorithms may limit students’ capacity to develop independent judgment, initiative, and creativity [32,38,45]. In the long term, there is a risk of generating functional dependency that could undermine essential skills such as reflection, analysis, and decision-making without automated assistance [28,42,49]. If not properly addressed, this phenomenon could produce new forms of inequality—not in access to technology, but in the ability to use it wisely.

Another key challenge lies in ethics and transparency. As AI becomes more involved in decisive aspects of the educational process, it will be essential to ensure that its recommendations are understandable, impartial, and adapted to diverse sociocultural contexts [26,62,64]. The future of online education requires the development of regulatory frameworks that safeguard student privacy and ensure respect for cognitive diversity, while avoiding homogeneous or prescriptive learning models.

Moreover, the true potential of AI will be realized when these tools are embedded within pedagogical models that foster active student engagement. The use of intelligent tutors, immersive simulations, and adaptive platforms must be accompanied by methodologies that promote exploration, dialogue, and critical reflection. This convergence between AI and transformative pedagogy can help consolidate a more resilient, inclusive, and sustainable educational model.

Thus, Artificial intelligence represents a transformative force that will shape the future of self-directed online learning. Its impact will depend not only on how the tools are designed and implemented but also on the ethical, pedagogical, and human-centered approach that guides their use. The key lies in combining technological efficiency with holistic education so that the learner remains at the center of the learning process.

4. Discussion

The results presented in this study offer a more detailed perspective of the role played by artificial intelligence (AI) in self-directed learning processes. From an overall perspective, the evidence reinforces the idea that AI technologies can act as essential facilitators to enhance student autonomy, personalization, and motivation in these processes, but it also reveals limitations and challenges that must be carefully considered.

In line with previous research, such as those by MA et al. [20] and Younas et al. [21], the current findings show that AI tools, particularly conversational assistants, intelligent tutoring systems, and automated assessment platforms, significantly promote the personalization of learning. These technologies enable content adaptation [44,69], provide immediate feedback [26,37,76], perform accurate diagnostics [83], and assist in study planning [30,40,44], all of which are essential in self-directed learning, where the student assumes a central role in their learning process. Furthermore, previous studies have highlighted that interaction with these tools can increase motivation and the perception of competence [26], key factors in maintaining engagement in autonomous learning processes.

However, this study also reveals that the implementation of AI in these contexts still faces significant obstacles. One of the most relevant challenges is technological dependence, which can lead to cognitive overload [21,56], disconnect students from their natural learning process [69], or even cause frustration due to technical failures or inaccurate or biased responses from AI systems, such as “hallucinations” in the generated responses [60]. These limitations are reflected in previous research that points out that the reliability and quality of content generated by AI are critical aspects for effective adoption in education, especially in contexts where accuracy and knowledge validation are imperative [84].

From a theoretical perspective, these results reinforce the hypotheses that AI integration can, in fact, promote greater autonomy and self-management, but to maximize these benefits, a pedagogical design aligned with the technological capabilities and limitations is fundamental [20]. Pedagogy should focus on enriching experiences, promoting metacognitive skills, and fostering prompting strategies that allow students to interact effectively with AI tools, minimizing the risks of dependence or misuse. Training in prompting skills, in particular, emerges as a key factor to fully harness the potential of AI technologies [85].

Another relevant aspect is ethics and equity in the use of educational AI. The potential disparity in access to these technologies, along with concerns about privacy, data protection, and algorithmic biases, must be addressed more thoroughly to ensure that pedagogical innovation is aligned with principles of justice and social sustainability.

In online education, the integration of Artificial Intelligence is rapidly transforming the landscape of self-directed learning, positioning AI as a cornerstone in the evolution of digital education. As Zeeshan et al. [86] highlight, emerging technologies are fostering the development of sustainable, internet-based educational models, particularly through smart, connected learning environments that prepare students for future societal and labor market demands. This trend directly supports Sustainable Development Goal 4 by promoting inclusive, equitable, and quality education through digital means [87].

AI-driven platforms play a critical role in empowering learners to take control of their educational journeys. These systems offer personalized content, adaptive learning pathways, and intelligent feedback mechanisms that support deeper topic exploration, time management, and autonomy. As such, AI enhances key components of self-directed learning by not only facilitating access to specialized knowledge but also cultivating metacognitive skills and learner agency. Nevertheless, as Garshid and Asham [88] caution, these benefits are contingent upon equitable access to technology, the quality of digital resources, and the ongoing digital literacy of both learners and educators.

Still, the future of internet-based education with AI hinges on thoughtful design and ethical implementation. Qu et al. [89] emphasize that current AI integration in education remains nascent, calling for a broader interdisciplinary effort to advance its development. As Sharples [90] asserts, it is crucial that AI tools are designed not solely to optimize performance but to uphold human values, empower educators, and address diverse learner needs. This demands collaboration among experts in AI (both neural and symbolic), pedagogy, and learning sciences, in coordination with active educator participation, to ensure that the future of internet education is not only intelligent but also inclusive, ethical, and pedagogically sound.

4.1. Practical Implications and Future Lines of Research

The findings of this study have significant practical implications for educators, instructional designers, and educational institutions seeking to integrate artificial intelligence into self-directed learning (SDL) environments. The results suggest that effective implementation of AI tools requires not only strategic selection but also alignment with pedagogical goals that prioritize learner autonomy, self-regulation, and critical thinking. These competencies are essential in the evolving landscape of online education and must be explicitly supported by the technological solutions adopted.

To enhance the applicability of these findings, institutions are encouraged to adopt a learner-centered integration model, where AI is used not as a replacement for human facilitation, but as a support system that adapts to individual needs, provides real-time feedback, and encourages independent exploration. For example, conversational agents like ChatGPT can be leveraged for formative feedback and reflective questioning, while intelligent tutoring systems can guide learners through personalized learning pathways. To ensure effective application, continuous digital literacy training should be provided to both students and educators. This includes not only technical skills but also critical competencies such as prompt engineering, ethical use of AI, and interpretation of AI-generated content.

The use of AI in SDL should also be framed within active learning pedagogies, such as project-based learning, collaborative inquiry, and problem-solving frameworks. These models amplify the value of AI by situating it within meaningful, socially rich learning contexts, allowing students to engage in higher-order thinking while benefiting from adaptive technologies. For instance, combining AI-supported feedback with peer collaboration and metacognitive strategies can strengthen learners’ ability to regulate their own learning over time. However, it is important to acknowledge the potential tensions between individual autonomy and structured collaboration. In project-based learning, clearly defined goals and shared responsibilities may limit flexible, self-directed trajectories, while collaborative inquiry emphasizes social interdependence that can challenge individual control. In this context, AI can play a mediating role by offering personalized scaffolding, such as differentiated group roles, real-time progress monitoring, or reflective prompts, supporting both individual autonomy and collaborative engagement. These affordances position AI not only as a tool for personalization, but also as a dynamic facilitator of balance between learner agency and collective learning processes.

From a research perspective, there is a pressing need for longitudinal studies that assess the sustained impact of AI on SDL beyond short-term interventions or pilot experiences. Current literature tends to focus on immediate outcomes, leaving open questions about how AI influences learners’ motivation, metacognitive development, self-efficacy, and digital competence in the long term. Furthermore, individual variability must be more thoroughly explored. Future studies should examine how factors such as students’ prior digital readiness, cognitive styles, or sociocultural backgrounds mediate the effectiveness of AI in different educational settings.

It is also necessary to expand the geographic and disciplinary scope of AI-in-education research. Most studies have been concentrated in the field of higher education and in technologically advanced regions, which limits the generalizability of their findings. Comparative studies across educational levels (primary, secondary, vocational), disciplines (STEM, humanities, arts), and cultural contexts can provide a more comprehensive understanding of how AI can be tailored to diverse learning environments. Additionally, future research should investigate the ethical and regulatory frameworks needed to safeguard equity, transparency, and learner agency in AI-mediated education.

Therefore, for artificial intelligence to meaningfully transform self-directed learning, its implementation must be carefully planned, student-centered, and responsive to the diversity of educational contexts. Only through a reflective, inclusive adoption aligned with principles of equity and quality will it be possible to fully harness its transformative potential within contemporary educational systems.

4.2. Limitations

One of the limitations of this study concerns the selection of bibliographic sources, as the search was restricted to the Scopus and Web of Science databases. Although both offer broad interdisciplinary coverage and high indexing standards, this methodological choice may have excluded valuable research from more specialized fields such as education and psychology, including studies indexed in ERIC or PsycINFO. Given that this study’s focus involves pedagogical dimensions such as student autonomy, motivation, and self-regulation, it is likely that relevant contributions on these topics were overlooked. Therefore, it is recommended that future reviews consider incorporating specialized databases, which would allow for a richer and more representative understanding of the field of self-directed learning supported by artificial intelligence technologies.

Another limitation is the variability in the quality of the studies reviewed and the lack of sufficient data to conclusively evaluate the effectiveness of AI tools, as well as the absence of longitudinal research that would enable analysis of the long-term impact of these technologies on self-directed learning. It is recommended that longitudinal and multicenter studies should be conducted to obtain solid evidence on the impact of AI technologies on self-directed learning, and that pedagogical guidelines should be developed that optimize their integration in various educational settings and promote greater effectiveness and sustainability of the learning process.

5. Conclusions

This study offers an integrative view of how artificial intelligence (AI) is currently being used to support self-directed learning (SDL) and reflects on its pedagogical, ethical, and practical implications. The findings show that a wide range of AI tools are being employed in SDL contexts, including conversational agents (e.g., ChatGPT), educational assistants (e.g., Khanmigo, Duolingo AI), intelligent tutoring systems, writing support tools, and diagnostic platforms. These applications serve varied purposes, from content generation and personalized feedback to performance tracking and guided exploration, demonstrating AI’s growing relevance across different dimensions of the learning process.

AI is being implemented primarily in higher education and health sciences, where students benefit from more autonomy and are often required to engage in complex problem-solving and continuous learning. Other emerging contexts include language learning, continuing education, and professional development. These settings share a demand for flexible, adaptive, and personalized instruction, making them fertile ground for the integration of AI to foster SDL.

In terms of learning processes, AI contributes most significantly to feedback, tutoring, assessment, and learning planning. It enables learners to receive timely guidance, set goals, organize tasks, and monitor their progress, core elements of effective SDL. Moreover, AI supports exploratory learning and the practical application of knowledge, particularly through adaptive simulations and intelligent recommendations, which can increase learner engagement and self-efficacy.

Despite these advantages, the study also identifies challenges that must be critically addressed. Ethical concerns such as data privacy, algorithmic bias, and academic integrity emerge as central issues. There are also pedagogical concerns, including the risk of reducing human interaction, fostering over-reliance on automation, and diminishing students’ critical thinking and initiative. Technical barriers such as inaccurate outputs, cognitive overload, and a lack of prompting literacy among users further complicate the effective use of these tools. These findings underscore the importance of thoughtful integration strategies that emphasize pedagogical soundness, transparency, and inclusiveness.

Looking ahead, AI has the potential to reshape the future of online education by enabling more personalized, flexible, and autonomous learning environments. However, this potential will only be realized if implementation efforts are guided by ethical foresight and grounded in robust pedagogical frameworks. Rather than replacing human educators, AI should be leveraged to enhance their role and empower learners to become active agents in their educational journeys. The challenge moving forward is to ensure that technological innovation goes hand in hand with a commitment to equity, quality, and the holistic development of learners.