The Integration of AI and Metaverse in Education: A Systematic Literature Review

Abstract

The use of the metaverse in educational environments has grown significantly in recent years, particularly following the shift of major tech companies towards virtual worlds and immersive technologies. Virtual reality and augmented reality technologies are employed to construct immersive learning environments. The metaverse is generally understood as a vast digital ecosystem or virtual space, facilitating the transition of individuals from physical to virtual environments, and is applicable to educational domains where practical experiments are challenging or fraught with risks, such as space exploration, chemical experimentation, and flight simulation training. In addition, the integration of artificial intelligence with the metaverse within educational contexts has significantly enriched the learning environment, giving rise to AI-driven teaching systems tailored to each student’s individual pace and learning modalities. As a result, a number of research articles have been conducted to explore the applications of the metaverse and artificial intelligence in education. This paper provides a systematic literature review following the PRISMA methodology to analyze and investigate the significance and impact of the metaverse in education, with a specific focus on the integration of AI with the metaverse. We address inquiries regarding the applications, challenges, academic disciplines, and effects of integrating AI and the metaverse in education that have not yet been explored in most research articles. Additionally, we study the AI techniques used in the metaverse in education and their roles. The review affirms that the integration of the metaverse in education, with the utilization of AI applications, will enrich education by improving students’ understanding and comprehension across diverse academic disciplines.

1. Introduction

The metaverse is widely regarded as the next evolution of the Internet. It will transform what is currently a two-dimensional web to a three-dimensional, immersive digital space. From a scientific perspective, there is no unified definition or unified agreement clarifying the exact structure the metaverse will take [1]. However, the most prevalent definition in the literature describes the metaverse as “a three-dimensional online environment in which users represented by avatars interact with each other in virtual spaces decoupled from the real physical world” [1,2,3,4]. Other studies offer more in-depth definitions; for example, ref. [5] defines the metaverse as “a world in which virtual and physical realities interact and co-evolve, and in which social, economic, and cultural activities are carried out to generate value”.

From a technological perspective, the metaverse is expected to include and integrate many contemporary emerging technologies, such as virtual reality (VR), augmented reality (AR), mixed reality (MR), blockchain, digital twin, artificial intelligence (AI), and the Internet of Things (IoT), among others [6]. The success of the metaverse will largely depend on the extent to which it can deliver high levels of immersion, fostering a sense of presence to make virtual interactions feel real and meaningful. The immersion will primarily be achieved through VR technologies, along with MR and AR. In recent years, these technologies, which are grouped under the umbrella of extended reality (XR), have evolved rapidly. The development of XR head-mounted displays (HMDs) has been driven by the integration of multiple sensors, computational capabilities, and advanced tracking features, enabling hand tracking, eye tracking, and facial tracking.

Constructing a metaverse that will replicate the complexities of the real world is a challenging task, requiring robust theoretical research at the framework level. Designing a well-defined framework prior to implementation is vital to establish a solid foundation for the system’s architecture, thereby guiding the development process. This framework plays a critical role by identifying the key components, interactions, and data flows necessary to integrate AI, virtual environments, and user interfaces effectively. One of the most discussed and adopted frameworks is that developed by Jon Radoff [2,7,8,9]. As depicted in Figure 1a, the proposed framework comprises seven layers emphasizing both technological and experiential aspects of the metaverse: infrastructure, human interface, decentralization, spatial computing, creator economy, discovery and experiences. However, ref. [5] proposed a framework (see Figure 1b) categorizing the metaverse into two key areas, technology and ecosystem, detailing how each interrelates to support a unified virtual environment. In terms of the technology aspect, eight key enabling technologies could be identified, including XR, AI, blockchain, and edge computing, which facilitate user interactions and enhance experiential aspects of the metaverse. The ecosystem aspect focuses on user-centric factors, such as content creation, the virtual economy, and social norms, which must align with pre-existing real-world regulations.

It is essential to distinguish the metaverse from standalone VR platforms and applications. While VR is a key component, the metaverse represents a broader integration of technologies, including VR, AR, MR, digital twins, IoT, etc. Together, these technologies enable a seamless fusion of virtual and physical experiences. Moreover, unlike standalone virtual worlds, the metaverse is envisioned as an interconnected ecosystem of virtual spaces, allowing for interoperability and the continuity of user experiences across different platforms [10].

1.1. Metaverse and Education

The utilization of the metaverse in education has attracted significant attention from researchers, especially given the constraints of traditional 2D teaching methods, as highlighted during the COVID-19 pandemic [11]. The key advantage of the metaverse from the perspective of educators is its capacity to enhance learning by providing immersive, interactive, and collaborative experiences. Immersive experiences will allow learners to explore complex concepts in a hands-on manner, enhancing their understanding [12]. Moreover, interactive elements, such as 3D models and simulations, will make learning more enjoyable, effective, and participatory. These interactive elements will allow for the exploration of real-world scenarios in multiple disciplines. In addition, teachers will be able to improve their skills using augmented reality (AR) tools in metaverse environments, thereby making training more realistic and engaging [13]. The metaverse offers numerous benefits for education. It provides wide-reaching access to knowledge, allowing students from various parts of the world to interact and learn from each other through avatars in a virtual space. Students can engage in experiential learning, exploring and practicing complex, authentic tasks that may be unattainable in the physical world. The metaverse fosters interaction and collaboration with individuals beyond their usual reach and personalizes education to address individual learner needs [6]. Moreover, the metaverse can lead to the creation of virtual universities, which provide highly active and collaborative environments for learning. This can enhance the educational experience by offering more interactive and engaging platforms compared to traditional online learning tools. The potential for creating virtual environments where students can engage in activities such as orientations, campus tours, and even internships further underscores the versatility and utility of the metaverse in education [14].

The Acceleration Studies Foundation (ASF) categorizes metaverse environments into four distinct types for the purpose of education, namely, augmented reality, lifelogging, mirror worlds, and virtual worlds [2,14]. Each category differs in terms of how the digital technologies interact with physical and virtual spaces. Hence, each version of the metaverse has unique potential to transform educational experiences, catering to different aspects of interaction, personalization, and immersion. Augmented reality describes technology that overlays virtual objects or information over the physical world. It can enhance learning by providing interactive, context-sensitive educational content. Using smart devices, lifelogging involves tracking and recording personal data, including activities, emotions, or behaviors. In education, this can provide personalized feedback and insights into students’ learning patterns. Mirror worlds create a virtual representation of the real world, considered to be reality’s digital twin. The digital twin replicates real-world spaces but adds educational functionality, such as virtual laboratories or digital classrooms. These virtual worlds are entirely simulated environments in which users can interact using avatars.

1.2. AI Within Metaverse for Education

The fusion of the metaverse and AI technologies promises to revolutionize how we teach and learn. The study in [15] provides a thorough classification and projection of algorithms informing AI techniques, and indicates that the learning process could be dramatically improved by integrating education-focused AI within the metaverse. Such integration would enable the development of immersive, personalized, and interactive learning environments to meet the varying needs of students. In particular, it would be possible to adapt course material according to each student’s performance. AI in education has shown significant promise as an instrument for personalizing learning experiences and promoting active participation. Supervised learning algorithms such as decision trees and support vector machines (SVM) could help develop adaptive learning systems in which course material is adjusted according to individual students’ progress, e.g., AI-based intelligent tutoring systems [16]. These systems can provide real-time feedback, adapt content delivery, and enable learning tailored to each student’s pace. AI can also assist with analysing large datasets, identifying patterns in students’ behavior, and improving learning outcomes. Unsupervised learning techniques, such as K-means clustering, can uncover correlations between students’ learning styles and performance. This then facilitates targeted interventions for those struggling, or excelling beyond the standard curriculum [17]. Reinforcement learning models (like deep Q-networks) can dynamically adjust learning content based on students’ interactions, ensuring an engaging and responsive learning experience. AI systems can further monitor students’ emotional responses or engagement levels, using these data to provide additional support or increase the challenge as required [17]. In addition, AI-powered NLP tools, such as transformers, recurrent neural networks (RNN), and convolutional neural networks (CNN), can further enhance the metaverse by improving communication between students and educational resources. These tools can enable voice commands, translations, or context-aware learning activities within virtual spaces [17]. Thus, it is vital to understand the various AI approaches available for use in the metaverse for educational purposes.

1.3. Contribution

Our study involved a systematic review following PRISMA methodology aiming to analyze both the significance and impact of integrating AI within the metaverse in education. This study pays special attention to technical aspects. We evaluated 474 research publications based on predetermined criteria to address the following research questions:

1.

What disciplines are covered by the AI-driven metaverse in education?

2.

How does the integration of AI in the metaverse affect student engagement, motivation, and participation in educational activities?

3.

What AI techniques are utilized in the metaverse for education?

4.

What are the challenges of utilizing AI in the metaverse for education?

Our systematic review was motivated by several key factors. First, knowledge regarding the interdisciplinary scope of the AI-based metaverse in education is limited, warranting an exploration of the associated fields and disciplines. Understanding the disciplines influenced by the integration of AI within the metaverse will also provide valuable insights to stakeholders in the education and technology sectors. Second, our study aims to explore how the AI-based metaverse, particularly the AI-based virtual learning environment, influences students’ engagement, motivation, participation, and performance. Given that AI has shown promise in terms of personalizing learning pathways and promoting active participation, it is essential to evaluate its effectiveness in the context of the metaverse. Third, since AI encompasses abundant technologies, from natural language processing to computer vision, it is vital to identify the AI techniques used in the metaverse for education. This will help us to understand the AI methods and techniques that are most applicable and beneficial in this context. Finally, while the potential benefits of AI in the metaverse are considerable, significant challenges, including technical, ethical, and pedagogical challenges, remain. Recognizing these challenges is crucial for future research to ensure the accurate integration of AI in the metaverse for the purpose of augmenting education. Therefore, this systematic review will provide a comprehensive overview of the disciplines, impacts, techniques, and challenges associated with the integration of AI in the metaverse for education.

The remainder of the paper is organized as follows. Section 2 presents a review of the literature on the metaverse and AI for education, examining systematic review studies specifically. Then, we detail the different phases of our research methodology in Section 3. In Section 4, we address the four research questions by conducting a comprehensive analysis of the eligible papers. Section 5 comprises the discussion and includes a critical evaluation of key findings and recommendations according to the challenges identified. Finally, we conclude the paper in Section 7.

2. Background

Recent research has affirmed that the metaverse is currently transforming the field of education, and that AI is playing a crucial role [16,18,19]. In this section, we examine most of the recent relevant works related to the use of the metaverse and AI in education.

2.1. Metaverse and AI in Education

Metaverse in Education. Some studies conducted qualitative research concerning the adoption and integration of the metaverse into higher education [11,20,21], while others focused primarily on its potential, opportunities, and challenges. Refs. [22,23] examined the benefits of using the metaverse encompassing interactive learning and student engagement, along with the challenges associated with infrastructure, security, and ethical considerations. Ref. [6] provided insights into the practical applications of metaverse technologies and the related opportunities, along with challenges including ethical considerations and confidentiality. Ref. [17] discussed four potential applications of the metaverse in education, highlighting future research topics such as design frameworks for educational purposes, guiding rules and principles, and the cognitive and non-cognitive impacts. Ref. [2] classified the metaverse into four types based on two axes: first, augmentation versus simulation, and second, intimate versus external. The characteristics of the four types are described with examples found in educational environments, considering both the benefits and drawbacks.

Moreover, the comprehensive bibliometric analysis found in [24] explored the evolution of research into the use of the metaverse in education, emphasizing trends, academic production, and key concepts in scientific publications.

AI within Metaverse in Education. However, only a small number of works examining AI within the metaverse in relation to education remain. Ref. [18] provided a clear definition of the metaverse, discussing its potential applications and research issues in education from an AI perspective, highlighting the role played by AI and suggesting new roles to enhance learning experiences such as NPC tutors, tutees, or peers. Ref. [16] focused on the role, techniques, and applications of AI, discussing the potential of integrating AI within the metaverse to revolutionize the current education system, exploring a number of applications of AI within the metaverse, such as (1) personalized learning; (2) intelligent virtual assistants; (3) adaptive assessment and feedback; (4) collaborative learning; (5) intelligent content creation; (6) intelligent tutoring systems; and (7) data analytics for educational insights. Ref. [19] offered a detailed review of the use of AI and 6G to create immersive experiences in the metaverse, covering advancements in computer vision, machine learning paradigms, and wireless communication technologies. The study explored the various paradigms of machine learning, focusing on self-learning as the most extensively explored in relation to the metaverse. The paper also discussed how AI within metaverse can revolutionize healthcare education by creating immersive and dynamic scenarios that replicate with precision, and in detail, a diverse range of medical situations. Ref. [25] provided a comprehensive analysis of integrating VR, AR, AI, and big data into a unified educational framework within the metaverse. Through a case study of urban scenario simulations for education, it highlighted how AI contributes through behavioral analysis, where machine learning models evaluate student interactions within virtual spaces, and by dynamically adjusting content difficulty to optimize individual learning paths. In addition, the study discussed how the integration of cloud and edge processing enables real-time analysis, ensuring that educational content adapts to students’ progress. Ref. [26] conducted a qualitative and quantitative research on the effectiveness of a smart education model within the metaverse focused on enhancing learning outcomes for college English students at Zhejiang Open University. The model integrated AI-enabled assessment alongside smart pedagogy, a smart teaching environment, and multimodal teaching resources. Over a one-semester experiment with 60 students, AI played a central role in providing personalized and formative assessments, dynamically adapting to individual learning needs. Furthermore, the study highlighted that AI-based assessment was the most influential component, with the highest levels of student satisfaction, underscoring its critical impact in driving the effectiveness of the smart education model.

2.2. Related Systematic Reviews

Several systematic literature reviews have explored the metaverse in education. Ref. [27] reviewed fifty highly cited articles, summarizing their research questions, methodologies, findings, and innovative applications such as intelligent transportation systems, AR for problem-solving, and virtual platforms for collaborative learning. It also highlighted uses in schools and healthcare education. Similarly, ref. [28] examined the metaverse’s potential benefits and drawbacks using content and bibliometric analysis. The study discussed trends, educational fields, digital identities, and the evolution and limitations of the metaverse in education. One of the study’s findings was that 53% of current metaverse studies employed technology for natural science, mathematics, and engineering. A review in [29] explored the perceived usefulness of the metaverse and its ease of use in educational contexts, encompassing numerous domains from 2008 to 2022. It demonstrated the vital role that perceptions play (including usefulness and ease of use) in metaverse acceptance, because they directly influence student and teacher decisions to use or reject this technology. In addition, the review found that perceived usefulness had a significant positive relationship with the intention to use an educational AI chatbot. Furthermore, ref. [30] reviewed seventy-three studies and proposed a framework to adopt the metaverse in education. Using six research questions, the authors analyzed publication trends, applications, dominant themes, challenges, and factors influencing metaverse adoption. They identified gaps in existing research and proposed twenty-seven new theoretical directions, including foundations, methodologies, and contextual challenges for implementing the metaverse in education. Ref. [31] examined forty-one research articles to assess the acceptance and adoption of the metaverse in education. The study identified the Technology Acceptance Model (TAM) as a key framework for understanding these aspects, with most research focusing on university students in China, Taiwan, and the USA. The authors also analyzed key research fields, including categories in the education sector, primary research methods, major disciplines, and educational levels. The authors of [32] considered various technological and pedagogical features, learning tasks, and design approaches in educational immersive VR (IVR) implementation. They categorized the design of IVR learning into five distinct approaches, revealing differing levels of interactivity and openness. They suggested that future research should explore the links between design features, learning approaches, and outcomes, so as to improve the understanding of the educational impact of immersive VR. Ref. [33] focused on design elements, learning content, and the use of immersive technologies in higher education. The study highlighted the significant potential of VR in eighteen areas of higher education but noted that few studies consider the role of learning theories in developing VR applications, which are crucial for enhancing student learning. The researchers also observed that VR is primarily used in experimental settings rather than regular classrooms, with an emphasis on ease of use over learning benefits. The study by [34] discussed the contributors, scientific collaborations, and thematic structure of metaverse research in education. The study found that most contributors come from various countries and institutions, with a notable concentration of papers from the USA, Japan, South Korea, and other parts of Asia. The researchers identified key themes in the metaverse for education, including (1) virtual sports; (2) health and medical learning simulations; (3) art education in 3D environments; (4) digital STEM labs; (5) immersive language learning; (6) social skills development for children with autism; (7) game-based learning; and (8) learner experience studies. The study by [14] reviewed articles from 2021 to 2022, establishing a systematic analysis of how metaverse technology is used in education. The research questions focused on the types of metaverse technologies employed, evolving metaverse trends and the challenges of using metaverse technologies in education, particularly in Indonesia. The paper also identified leading authors and their contributions to the field, so indicating the impact of the most influential works.

The study by [35] conducted a systematic analysis of articles, examining aspects such as subject, journal, year, research method, study group, sample, data analysis and collection methods, field, country, conclusions, and suggestions for future research. The study found that the metaverse enhances student engagement and communication but also identified challenges, including the difficulty of creating content for these technologies and the risk of students becoming overly immersed in virtual worlds. Additionally, it noted a growing interest in the metaverse, particularly its impact on education and computer use, with a focus on individuals aged 18–40. The study by [36] provided a systematic review of the factors encouraging the use of the metaverse within learning environments, organizing them into four main groups: (1) psychological and motivational; (2) quality; (3) social; and (4) inhibiting factors. The paper offered also a framework to guide future research on this topic. Ref. [13] discussed the importance of the metaverse in education, including how it can improve learning, skill development, and training using extended reality (XR) and the Internet of Everything (IoE) technologies. It provided a detailed review of educational applications within the metaverse, analyzing the technologies built on XR, IoE, big data, and AI.

Table 1. Survey of the systematic review articles.

Reference	Articles/Time Span	Focus
[13]	182 out of 3143/No time constraints	How the metaverse can improve education through the use of extended reality and the Internet of Everything technologies.
[36]	9 out of 85/up to December 2022	Metaverse adoption in education. The reasons individuals may wish to use the metaverse for education.
[35]	33 out of 603/No time constaints	How the metaverse influences its users.
[27]	50 out of 311/Between January 2015 and April 2023	Use of the metaverse for educational purposes, along with challenges and risks.
[28]	77 out of 77 /Up to December 2002	Research trends, focus, and limitations in the educational metaverse.
[29]	5 out of 357/Up to 14 November 2022	Acceptability of the use of metaverse in education over 16 years.
[30]	73 out of 365/Between December 2012 and May 2023	Applications, challenges, themes of the research, and future perspectives of the metaverse in education.
[31]	41 out of 90/Up to May 2022	Acceptability of the use of the metaverse in education.
[32]	219 out of 1728 (exclude papers with less than 5 citations per year)/Between 2016 and 2021	Design of learning in immersive VR.
[33]	80 out of 2646/Between 2016 and 2018	Design elements of existing research dedicated to the application of VR in higher education.
[34]	310 out of 413/Between 2004 and 2022	The identity of those undertaking research concerning the metaverse for education, how they work together, and the main focus of their topics.
[14]	42 out of 1305/2012–2022	The prevalence and current status of metaverse technology within the educational sector.

summarizes the above systematic reviews, detailing the number of eligible papers out of the number of the retrieved articles, time span, and the focus of each study. The number of retrieved articles was obtained upon the search phase and does not include duplicated papers.

2.3. Research Gap

As summarized in Table 1, the study found no systematic review focusing on the use of AI within the metaverse for education. The keywords of related works revealed a lack of any research into “education”, “metaverse” and “AI” subjects except [13]. The keywords in [33] explicitly exclude AI by using a NOT operation to avoid articles reporting on AI without including the human learning context. In addition, most papers failed to address the issue of AI, either superficially or in depth [14,27,30,31,32,33,34,35]. These reviews mainly focus on the use of the metaverse to make learning more effective and engaging, stressing aspects such as (1) the adoption of the metaverse [30,31]; (2) the main disciplines covered by the metaverse [31]; (3) the design of immersive learning content [32,33]; (4) the contributors to the metaverse [34]; (5) metaverse technologies [14]; (6) future research [30,35] and (7) challenges [27,30,35].

Nevertheless, we also found only a limited number of systematic reviews examining AI from a specific perspective [13,28,29]. Ref. [13] focused primarily on IoE and XR technologies, examining how AI could be integrated with these technologies to support education. Furthermore, the review in [28] discussed the role of intelligent non-player characters (NPCs) within the metaverse, as well as the application of a number of AI techniques (i.e., neuro-symbolic AI, convolutional neural networks, machine learning) in the metaverse for education. In addition, ref. [29] considered AI as part of an analysis of the perceived usefulness of the metaverse, with the authors arguing that this can be enhanced by the use of chatbots.

While it is conceivable that, in the future, the metaverse and AI will enhance the advancement of education, their use currently remains at an early stage of development. We found this area currently largely unexplored in systematic literature reviews and therefore aim to fill this gap in the existing literature, which could also encourage future studies on this subject.

3. Data Sources and Search Strategies

We conducted the present systematic literature review using the guidelines of the Preferred Reporting Items for Systematic reviews and Meta-Analyses, PRISMA [37]. The main purpose of this method is to grant a certain level of transparency on the conduct of the reported systematic review elaboration, and therefore ensure the trustworthiness of the ensuing results. In addition, it facilitates replication of the review. It includes four steps: (1) identification; (2) screening; (3) eligibility; and (4) inclusion [38].

3.1. Keyword Occurrences

To examine the relationship between keywords in the areas of AI and the metaverse in education, we applied VOSviewer software, version 1.6.8, to articles containing keywords as demonstrated in Figure 2. VOSviewer is a software application for visualizing and studying bibliometric networks, particularly those generated from databases of the scientific literature [39,40]. In order to investigate the searched keywords, we merged the results of the searches of the Web of Science and Scopus databases. Figure 2 visualizes the co-occurrence of keywords in the selected publications with circle sizes and names proportionate to the number of occurrences within the articles. In addition, the link between the keywords reveals their relationship, while the distance between any two keywords represents the strength of the link [41,42]. For example, “metaverse”, “artificial intelligence”, and “virtual reality” are placed close to each other, with the number of occurrences being 207, 124, and 114, respectively. The “education” keyword occurred ninety-three times, both singly and combined with other keywords such as “engineering education”, “medical education”, “education metaverse”, “design education”, and “education and metaverse”. The keywords “metaverse” and “artificial intelligence” are linked with the keyword “education” and its combined keywords. Figure 2 also shows the keywords “metaverse”, “metaverses”, “virtual reality”, “artificial intelligence”, and “augmented reality” as the trending research areas. There are four main clusters, with the violet cluster centered around “metaverse” and “metaverses”, which are linked with “artificial intelligence”, “virtual reality”, “education”, and “engineering education”. The brown cluster shows the research keyword “virtual reality”, which also relates to “metaverse” and “artificial intelligence”. The orange cluster represents the “artificial intelligence” keyword linked with the keywords “metaverse”, “virtual reality”, “education”, and “machine learning”. Finally, the turquoise cluster consists of the centered keyword “AR” with related keywords “mixed reality”, “digital twin”, “metaverse”, “artificial intelligence”, and “education”.

3.2. Derivation and Search Term Composition

This current work is based on three principal terms, namely “Metaverse”, “Education” and “Artificial Intelligence”. We considered these terms to cover relevant academic papers related to the use of AI techniques in the metaverse for education. As “Artificial Intelligence” is an umbrella term, it also includes other expressions, which led us to extend the search list by selecting the most relevant keywords related to the basic term “Artificial intelligence”. This led to the following terms: “Machine learning”, “Deep learning”, “NLP”, and “Chatbot”. In the same way, we extended the term “Education” to “Teaching”, “training”, “Student”, and “University”. The inclusion of “Learning” as a synonym of “Education” led to excessive noise, due to it appearing in the text of articles without referring to its intrinsic meaning, but rather to machine learning methods. This resulted in the potential for articles unrelated to education to bias the search result. However, when it came to the third main term, “Metaverse”, we did not include its related terms (i.e., “Augmented reality”, “Virtual reality”, “Mixed reality”, and “Extended reality”) due to confusion concerning the differences between all these related concepts. We found that, despite the large number of definitions of the metaverse [43,44,45] and the interchangeable use of these technologies, there remain some important differences, primarily that the metaverse is a user-created online decentralized digital environment, in which participants interact through self-created avatars. To avoid confusion, we decided to focus only on the concept of the metaverse. In addition, it should be noted that a large search term list tends to be problematic, since it generates a huge result containing records that are not necessarily relevant, and is highly time- and effort-consuming for PRISMA screening stages. By contrast, a succinct search term list leads to results missing some records relevant to the review. This led us to form the final search query by combining the previously selected terms.

3.3. Data Search and Selection

The search process for this systematic review was conducted for the last decade and ended on 22 August 2024. We selected Web of Science (WoS) and Scopus as research databases for articles, as these are the most widely used and authoritative databases of research publications and citations. Among the search results, we considered journals and conference papers to cover the maximum studies of the targeted review topic [46]. We used the following search phrase in Scopus: “(Metaverse) AND (Education or Student or Teaching or Training or University) AND (“Artificial Intelligence” or ”Machine Learning” or ”Deep Learning” or NLP or Chatbot)”. We also used the same search phrase in Web of Science. Note that the search involved the title, abstract, and keywords. In the following, to justify the inclusion studies process, we describe the review reporting using PRISMA stages.

3.4. Identification

The research publication identification related to the review topics is the initial stage of the PRISMA protocol. We collected 151 articles and 314 articles in WoS and Scopus, respectively, resulting in a total of 465 articles.

3.5. Screening

We then screened the initial search result of articles as follows. First, we deleted 113 duplicate articles to obtain 352 articles. Second, we excluded 255 articles due to their titles and abstracts being unrelated to the scope of the review. Third, we undertook a manual search to retrieve references that may have been missing from the initial search, so as to ensure that we included all possible relevant articles and to reduce any bias. Fourth, we assessed the inclusion of the resulting nine additional related papers in the same manner as the records identified from the searching databases.

3.6. Eligibility

After the screening step, we assessed the remaining full-text articles for eligibility. Among the remaining 97 articles, 11 lacked accessibility to the full text and the remaining 86 were full-text articles assessed in terms of eligibility for the study according to the inclusion and exclusion criteria (see

Table 2. Inclusion and exclusion criteria.

Inclusion Criteria	Exclusion Criteria
Includes metaverse and AI techniques. Responds to at least one of the research questions. Full text available. Written in the English language. Journal or conference article.	Does not involve metaverse and AI techniques. Is not related to research questions. Article was not accessible. Article was published in language other than English. Report, book, chapter, lecture notes, thesis, proceedings, or dissertation.

). We determined that 33 articles were not related to any of the research questions stated in Section and did not focus on education or AI techniques. To reduce bias as much as possible, we proceeded to a manual search through reference lists and citations to identify any further relevant studies. We collected nine more articles for eligibility and we include them in the “studies included in review”. All articles were independently assessed by two reviewers and selected according to the eligibility criteria. At the beginning, the articles are assigned equally to four readers (F.E.A., M.B., B.I., and H.M.). Then, the readers exchanged the articles for a second review. In case of disagreement or uncertainty, a third review was carried out by K.A. The obtained papers from the database search and additional papers from the reference search were examined again by all the reviewers regarding the research questions. In this step, information was extracted and encoded using Microsoft Excel.

3.7. Inclusion

In this final step, we subtract the number of records excluded during the eligibility review from the total number of articles reviewed for eligibility. We obtained 62 articles to considered in this review.

The details of the filtering process of the identified articles are illustrated in the PRISMA flow diagram (see Figure 3).

4. Results and Discussion

To obtain responses to four research questions, the present systematic review looked at sixty-two research articles on metaverse-based AI techniques for education, as discussed below.

4.1. Rq1: What Disciplines Are Covered by the AI-Driven Metaverse in Education?

Through this question, we intend to present a clear vision of the education disciplines using metaverse-based AI techniques explored in the current literature, as well as the potential disciplines for future investigations to encourage the adoption of such techniques. Firstly, we determined that metaverse-based AI can play a crucial role in modern education, particularly due to the current rapid advances in AI techniques. Secondly, we established that metaverse technology is being introduced into several educational disciplines, including (1) medical education [47]; (2) aircraft maintenance training [48]; (3) tourism education [49]; (4) law [50]; (5) English language education [51]; (6) engineering education [52]; (7) Building Evacuation Training [53]; and (8) Chinese language education [26]. We also identified a number of educational disciplines that have not yet benefited from this technology, but have the potential to employ existent metaverse-based AI tools to adopt state-of-the-art techniques into their domain. These include (1) history; (2) languages other than English and Chinese; (3) geography; (4) biology; (5) chemistry; (6) astronomy; (7) physics; (8) civil engineering; (9) electrical engineering; (10) manufacturing engineering; (11) environmental studies; (12) journalism; (13) military sciences; and (14) geology. It should also be noted that each of these disciplines can be group together several other branches. For example, medical education contains many subfields capable of integrating metaverse technology to establish more intelligent healthcare learning. These include (1) dentistry; (2) orthopedics; (3) cardiology; (4) genetics; (5) neurology; (6) obstetrics; (7) ophthalmology; (8) and surgery training.

To obtain a response to this research question, we reviewed the total number of research publications gathered in the final PRISMA step. Figure 4 shows the distribution of the reviewed papers in terms of their disciplines, revealing that the largest number (N = 30) conducted general studies of metaverse-based AI without specifying a specific educational discipline. These we categorized as General. The second most relevant set of papers considered medical education (N = 9), with fewer articles examining language education (N = 3) and educational games (N = 3). The remainder of the papers addressed other distinct disciplines.

As shown in Figure 4, many education disciplines remain unexamined, and we therefore propose that these should be investigated to address the benefits of the use of the metaverse driven by AI in education. Moreover, we consider that this could be achieved by adopting the findings and AI tools found in general studies within many specific disciplines by taking into consideration their individual specificities. We estimate that the coming years will present great potential and new possibilities to investigate other promising education disciplines. However, some may be less suitable, including philosophy and political education. This has highlighted that the use of the metaverse requires an educational field to possess certain features, including the need to master skills, known as a skill-based learning approach. This finding is due to the metaverse providing a more immersive and engaging educational experience in a three-dimensional virtual environment capable of mirroring a realistic situation to a wider number of learners and thus saving both time and money. It also offers the possibility to practice numerous skills and repeat them as often as desired, permitting the creation of a more immersive experience than a single experience in the real environment, i.e., medicine, sport, aviation training, and laboratory work. However, the disciplines with most concerns when it comes to the adoption of the metaverse are those that promote the following aspects: (1) teacher–student communication; (2) cooperative learning; (3) active student participation; (4) team discussion; (5) critical thinking and group competition; (6) group activities; (7) peer teaching; and (8) the exchange of knowledge. AI plays a major role in supporting the metaverse to reinforce these educational features through the use of the following: (1) machine learning [54]; (2) chatbots [51,55,56]; (3) recommender systems [57]; (4) intuitive interfaces [58]; (5) the generation of facial expressions [59]; (6) natural language processing [60,61]; (7) voice recognition [48,62]; (8) computational intelligence [60]; (9) reinforcement learning [53,63]; (10) intelligent agents (NPCs) [64]; (11) personalized services for learners [51]; and (12) decision-making [18]. Furthermore, integrating these AI techniques with the metaverse allows it to become extremely powerful.

4.2. Rq2: How Does the Integration of AI in the Metaverse Affect Student Engagement, Motivation, and Participation in Educational Activities?

This study found that the integration of AI and the metaverse in education has opened up new possibilities for enhancing learning experience and improving student outcomes. This emerging field (often referred to as the edu-metaverse) combines advanced technologies to create interactive and engaging virtual learning environments that simulate real-world experiences and provide personalized educational opportunities. There are a number of applications of the metaverse in education, such as, firstly, utilizing gamification for educational purposes, so converting educational practices to enjoyable games, and secondly, conducting experiments that are difficult to perform in reality in virtual wolds, e.g., hazardous chemical reactions and pilot training.

Several studies have discussed the impact of the metaverse in education on students’ engagement and motivation in relation to different fields of education and in different countries as well as cultures. Some researchers have developed tools to measure students’ engagement, motivation, and participation. In Taiwan, Human Intelligence (HI)-based Computational Intelligence (CI) and AI Fuzzy Markup Language (CI&AI-FML) methods have been proposed for academic individuals, with both teachers and students utilizing these to interact with the metaverse environment. Furthermore, the interaction of students with metaverse tools such as VR and AR showed their level of comprehension and engagement [60]. An additional metaverse-based collaborative learning interface was found to be capable of evaluating students’ interaction with the metaverse and AI. The study engaged twenty-one participants from distinguished universities in Thailand and Myanmar, heightening student’s engagement by personalizing the experience based on individual learning patterns and utilizing gamification tools [58].

This current review found that there has been a notable rapid proliferation of AI and metaverse integration within educational contexts. Several studies have examined the attitudes of students and faculty members to evaluate the application of the metaverse in education as an alternative to traditional approaches. A number of studies in Bahrain, Jordan, and the UAE examined data analysis tools, such as the TAM theoretical framework, suggesting that participants were optimistic when it came to the possible enrichments offered to education by the metaverse [65,66,67]. Furthermore, an AI-based simulation system can be designed to improve learners’ focus and immersion in a VR environment, focusing on evaluating learning behavior to allow learners to proceed to the next level of a course [68].

In addition, modern assistive technologies can improve student learning outcomes and increase educational accessibility, including students with learning difficulties. Thus, assistive tools in reading and writing can help such students convert from text to speech and from speech to text, while others like QTrobot are suited to children with autism, and Widex’s evoke is suited for use with hearing aids. In this context, a survey of 149 students with learning difficulties found that assistive learning tools enhanced both their reading skills and motivation [69,70]. In addition, advanced technologies have helped software developers offer education through gamification to enhance engagement, motivation, and participation. This was demonstrated in a game-based classroom (i.e., the VR escape game developed in the metaverse), which enhanced students’ capabilities for mobile learning, ensuring they were more engaged than in traditional classrooms, including a strong motivation to use smart devices to acquire science knowledge. This indicates that the metaverse offers an adaptive platform for game-based learning environments, and can also foster multilingual learning [28].

The pedagogical approach of AI- and metaverse-incorporated education is based on the concept of experiential learning, wherein students are engaged in exploratory activities, acquiring knowledge and skills in a virtual environment. Moreover, the implementation of gamification within an educational framework is inherently more appealing to learners, due to the merging of playful components, encouraging an intrinsic motivation towards learning [71]. It is possible to develop adaptive and personalized educational games with VR and AR based on the recognition of images, human interactions, speech, and intelligent agents, which can be utilized for the simulation of a human opponent and the manipulation of a game’s level of complexity [72]. In addition, language learning can be enriched by AI and metaverse environments, as revealed by a questionnaire on AI and the metaverse which was conducted with twenty-three language instructors, revealing the efficacy of language learning to be significantly more enhanced than that on Meet or Zoom [73].

VIRTUSIM is an immersive and adaptive virtual learning system incorporating AI and multi-agent technologies to offer an adaptive personalized education to students. The prototype system was implemented and tested for English language learning, resulting in a 30% improvement compared to traditional educational methods [74]. Studies have shown that 97.9% of elementary school students experienced the metaverse in almost everyday life, while preservice English teachers’ ability to create content in this virtual environment enhanced their teaching, thus demonstrating that the metaverse improves students’ active participation and engagement, as well as their interaction and motivation [51,75].

A study also found that the application of AI and the metaverse in medical education enhanced the training and comprehension of medical students. In addition, their use for emergency response training was able to guide responders to establish emergency protocols and strategies to help patients in an accurate and competent manner, enabling healthcare professionals to focus on decision-making and action-taking during an emergency. It also found that the metaverse can also enhance physical and mental health by providing virtual instructors [76]. The Artificial Intelligence, Innovation, and Society (AIIS) project identified requirements for immersive multiuser learning, including the constraints and opportunities of metaverse environments. The project proposed a first draft pedagogy model for a metaverse collaborative learning environment, enabling students to participate on a higher level than available in conventional education [77].

Studies have also determined that it is possible to perform medical treatments and pharmacological protocols on human avatars with symptoms as in real patients to facilitate medical students with relevant experience prior to practicing in the real world. In addition, the metaverse can offer medical students digital lessons, with the ability to dissect and to examine anatomical devices in detail, as well as monitor the evolution of pathological processes and simulate surgical procedures on human avatars [78]. Students’ motivation and engagement in the metaverse environment was measured quantitatively with a MANCOVA model (i.e., a statistical technique to analyze the relationship between multiple dependent and independent variables), which showed that the differences between the experimental group (metaverse education environment) and the control group (conventional education environment) were considered statistically significant. This strengthened the study’s conclusion that the integration of AI and the metaverse for the purposes of education can significantly improve learning outcomes, with 88% of students positively influenced by the metaverse learning technologies [26]. In a further study, the Attention, Relevance, Confidence, and Satisfaction (ARCS) model was used to evaluate students’ interactions with AR. The questionnaire demonstrated AR led to a 77% improvement in the students’ comprehension, with the use of AR applications in education proving to increase students’ motivation and participations through gamified learning experiences [79].

4.3. Rq3: What AI Techniques Are Utilized Within the Metaverse for Education?

This study determined that recent technological advancements, particularly the utilization of AI within the metaverse, have had a notable impact within the educational field. An analysis of N = 60 technical contributions from the total number of eligible papers identified AI techniques as possessing significant potential for enhancing educational practices through the application of metaverse techniques. These included (1) deep learning; (2) reinforcement learning; (3) supervised learning; (4) unsupervised learning; and (5) natural language processing. Prior to analysing these techniques, we provide some definitions in

Table 3. Definitions of AI techniques tackled by researchers for the metaverse and education.

Technique	Definition
Supervised Learning	This arises when tasks possess labeled data, and is a fundamental AI technique that supports pattern recognition and prediction. This method involves feeding datasets with known outcomes to AI systems to enable them to learn, with the chief objective being to identify patterns in the data to be utilized to translate input into the correct output. Speech recognition, recommendation systems, and picture classification are just a few of the fields in which supervised learning finds widespread use.
Unsupervised Learning	This differs from supervised learning due to the absence of labeled data. Thus, without any prior knowledge, AI systems examine unlabeled data to identify underlying structures or patterns. Unsupervised learning is frequently used in clustering and dimensionality reduction.
Reinforcement Learning	This is a branch of machine learning whose goal is the reinforcement of learning so as to maximize behavior in intricate settings. By using this method, AI systems acquire new skills by interacting with their surroundings and being rewarded or punished for their efforts. Reinforcement learning differs from supervised learning as it is based on experience rather than labeled data.
Deep Learning	This is a specific type of machine learning that models complex patterns and representations in the data using multi-layered neural networks. Applications such as speech recognition, picture identification, and natural language processing demonstrate the power of deep learning.
Natural Language Processing	NLP has made intuitive communication possible, revolutionizing human–computer interaction. It has enabled machines to comprehend, interpret, and produce human language and AI systems and to process, evaluate, and react to text or speech data in a manner that is similar to human language comprehension.

according to [15]:

An analysis of Figure 5 reveals some key insights regarding the usage of AI techniques in educational contexts within the metaverse. First, deep learning (51%) emerges as the AI technique most frequently employed, underscoring its crucial role in handling complex educational data and facilitating advanced features such as immersive simulations and personalized learning experiences. Additionally, the use of deep learning demonstrates its importance in evaluating student engagement through visual inputs and improving augmented reality experiences. References to natural language processing (23%) are also notable, probably reflecting the need for communication and interaction in educational environments, which has contributed to the development of applications like chatbots, virtual assistants, and automated feedback systems. Reinforcement learning (17%) focuses on tailoring learning pathways by adapting content based on users’ interactions in the metaverse. Supervised learning (7%) and unsupervised learning (2%) are less frequently discussed, indicating that while they play roles in predictive analytics and data clustering, they are not so central to the immersive and interactive features of the metaverse.

In summary, the prevalence of key techniques is as follows: deep learning (31 mentions) ranks first, followed by natural language processing (14 mentions), reinforcement learning (10 mentions), supervised learning (4 mentions), and finally, unsupervised learning (1 mention). Upon further evaluation, we chose to specifically investigate the AI algorithms integrated into the metaverse for education.

According to [15,80], it is useful to consider the following classification and projection of algorithms for AI techniques:

a

- Algorithms for Supervised Learning: Support vector machines (SVM), logistic regression, random forests, decision trees, linear regression, and fuzzy logic.

b

- Algorithms for Unsupervised Learning: Hierarchical clustering, principal component analysis (PCA), and K-means clustering.

c

- Algorithms for Reinforcement Learning: Q-learning, Deep Q network (DQN).

d

- Algorithms for Deep Learning: Long short-term memory (LSTM), recurrent neural networks (RNN), convolutional neural networks (CNN), and transformers.

e

- Natural Language Processing: Generative Pre-trained Transformer (GPT)

In

Table 4. Definitions of AI algorithms considered by researchers regarding the metaverse and education.

Algorithm	Definition	Reference
Linear regression	Linear regression is a statistical method for modeling the relationship between a dependent variable and one or more independent variables by fitting a linear equation to the observed data.	[81]
Logistic regression	Logistic regression is a statistical method used for binary classification, predicting the probability that a given input point belongs to a particular category using the logistic function.	[82]
Support vector machines	Support vector machines are supervised learning models used for classification and regression that find the hyperplane that best separates different classes in the feature space.	[83]
Decision trees	Decision trees are a non-parametric supervised learning method used for classification and regression by splitting the dataset into branches to form a tree-like model of decisions.	[84]
Random forests	Random forests are an ensemble learning method that constructs multiple decision trees during training and outputs the mode of the classes (classification) or mean prediction (regression) of the individual trees.	[85]
K-means clustering	K-means clustering is an unsupervised learning algorithm that partitions a dataset into K distinct clusters based on feature similarity, minimizing variance within each cluster.	[86]
Hierarchical clustering	Hierarchical clustering is an unsupervised clustering technique that seeks to build a hierarchy of clusters by either a bottom-up (agglomerative) or top-down (divisive) approach.	[87]
Principal component analysis (PCA)	PCA is a dimensionality reduction technique that transforms a dataset into a set of linearly uncorrelated variables (principal components) by identifying the directions of maximum variance.	[88]
Q-learning	Q-learning is a model-free reinforcement learning algorithm that seeks to learn the value of an action in a particular state, allowing an agent to learn how to optimally act in a given environment.	[89]
Deep Q network	DQN is a type of deep reinforcement learning algorithm that combines Q-learning with deep neural networks to approximate the Q-values, enabling the handling of high-dimensional state spaces.	[90]
Convolutional neural networks	CNNs are a class of deep learning models primarily used for processing grid-like data, such as images, by employing convolutional layers to capture spatial hierarchies and local patterns.	[91]
Recurrent neural networks	RNNs are a class of neural networks designed for processing sequences of data by using recurrent connections, allowing them to maintain information about previous inputs over time.	[92]
Long short-term memory (LSTM)	LSTMs are a special kind of RNN designed to learn long-term dependencies by utilizing memory cells that can maintain information for long periods, effectively addressing the vanishing gradient problem.	[93]
Fuzzy logic	Fuzzy logic is a form of many-valued logic that deals with reasoning that is approximate rather than precise, allowing for degrees of truth and enabling more flexible decision-making. Reference: Zadeh, L. A. (1965). “Fuzzy sets.” Information and Control.	[94]

, we have included some definitions of most AI algorithms discussed in this investigation.

In Figure 6, we can observe that convolutional neural networks (CNNs) were cited in 15 papers, indicating their widespread application in educational contexts for image recognition and interactive content creation. Principal component analysis (PCA) was mentioned in seven papers; this performs a dimensionality reduction to facilitate the analysis of complex educational data. Generative Pre-trained Transformer (GPT) was found in five papers, highlighting the use of NLP capabilities for chatbots, content generation, and personalized learning. Decision trees were mentioned in four papers, as they can assist educators in making data-informed instructional choices. Long short-term memory (LSTM) was also cited in four papers, as it is effective for managing sequential data and tasks such as tracking student progress. K-means clustering, which appeared in four papers, is beneficial as a tool for segmenting students’ needs or content as a way to tailor educational approaches. Logistic regression was mentioned in two papers, and is typically employed for binary classification tasks, such as predicting student success or dropout rates. Q-learning is a reinforcement learning approach and appeared in two papers, where it was suggested as useful for developing adaptive learning paths. Recurrent neural networks (RNNs) were cited in five papers, and are suitable for sequential prediction tasks that are relevant to educational analytics. Fuzzy logic was documented in two papers, and aids decision-making in uncertain contexts, potentially benefiting educational assessment. Computer vision was presented in five papers, indicating significant engagement with visual data for interactive learning. Support vector machines (SVM) were mentioned in one paper for classification purposes. Linear regression appeared in one paper, where it was discussed as useful for predicting educational outcomes based on various inputs. Deep Q network was not referenced in any papers, reflecting the limited exploration of advanced deep reinforcement learning techniques in education to date. Random forests were similarly not found in the reviewed papers, indicating a potential gap in the research in the area of ensemble learning methods. Hierarchical clustering was also absent, which may suggest there is a preference for simpler clustering techniques in the literature.

The gap between the number of papers discussing algorithms and the total number of source papers indicates that general mentions of techniques exceed specific algorithm applications. Consequently, the ranking of techniques can be updated as follows: deep learning (19 mentions) leads, followed by unsupervised learning (11 mentions), supervised learning (10 mentions), natural language processing (NLP) (5 mentions), computer vision (5 mentions), and finally, reinforcement learning (2 mentions) (see Figure 7).

Overall, this analysis illustrates a landscape in which deep learning and its related algorithms dominate AI applications for education within the metaverse, as driven by the demand for interactivity, adaptability, and user engagement.

Table 5. AI algorithms in the metaverse for education.

AI Algorithm	Purposes	Proposals
Linear Regression, Logistic Regression	Training a machine learning model on pre-processed data.	[95,96]
Support Vector Machines	Classification recognition for digital twin technology.	[97]
Decision Trees	(i) Forecasting how parameters in an algorithm will be correlated for machine learning; (ii) mimicking learner reactions that resemble those of a human.	[63,67,95,98]
K-Means Clustering	Improving the route for the transmission of 360-degree immersive VR football instructional films and determining the position for the closest proxy server for each participant.	[99,100]
Principal Component Analysis	(i) Integrating VR with machine learning to examine the accuracy of the trained algorithm; (ii) identifying high-level properties from raw eye tracker data.	[60,69,99]
Q-Learning, Hierarchical Reinforcement Learning	Increasing the intelligence of non-player characters.	[63]
Convonlutional Neural Networks	(i) Understanding speech executor; (ii) detecting emotion and offering recommendations regarding teaching methods; (iii) simulating how the nerve that supplies vision in humans interprets visual data; (iv) rating the second generation of web technology’s performance in the football classroom.	[48,63,68,96,100,101]
Recurrent Neural Networks	Understanding of context-aware speech in aircraft training and education.	[101]
Long Short-Term Memory	Connecting VR to machine learning.	[99]
Fuzzy Logic	(i) Incorporating human subjective cognition into the Item Response Theory (IRT) learning system; (ii) developing a better and captivating classroom for learners; (iii) providing a simple solution resembling GUI.	[60,102]
NLP	(i) Constructing a ubiquitous operating system model for the metaverse; (ii) examining the potential impact of chatbots on managerial training; (iii) studying how AI chatbots within virtual world settings affect preliminary English instructors; (iv) examining the impacts of using chatbots within the metaverse for education.	[51,56,103,104,105]

summarizes the AI algorithms and their proposals as previously identified in the eligible papers.

In response to Question 3, “What AI techniques are utilized within the metaverse for education?”, this study provides a scientific overview and classification. The findings, highlighted in Figure 5, Figure 6 and Figure 7, clarify that most of the reviewed papers concentrate on unsupervised learning, reinforcement learning, and deep learning.

4.4. Rq4: What Are the Challenges of Utilizing AI in the Metaverse for Education?

AI is crucial for enhancing the educational experience within the metaverse by providing tools including virtual teaching assistants, advanced language processing capabilities, and sophisticated learning outcome assessments. However, it is vital to continue to recognize the challenges faced when integrating AI within the metaverse, particularly as the metaverse development for education is currently at an early stage, and there remains a lack of comprehensive studies. The aim of this question is to prompt an in-depth investigation to establish the challenges associated with the use of AI within the metaverse for education, which opens a new axis for ongoing research in this field. The review of the sixty-two eligible papers identified several challenges, which we classified into five categories, discussed in detail below, and consisting of (1) technical challenges (N = 7); (2) algorithm bias (N = 3); (3) ethical and security concerns (N = 6); (4) integration and interoperability challenges (N = 4); and (5) computational and hardware challenges (N = 4). In Figure 8, we summarize the five key categories of challenges along with their respective percentages. The distribution highlights the complexity and multifaceted nature of AI deployment in this innovative educational environment.

4.4.1. Technical Challenges

The metaverse for education involves constant interactions between participants, which presents new scenarios for which AI needs to be adapted. However, current AI models are not yet specifically designed for such complex, interactive, and immersive virtual environments. Thus, it is challenging to create AI algorithms capable of effectively operating within the dynamic environments of the metaverse, due to requiring advanced algorithms capable of real-time adaptation and decision-making. A notable technical challenge is the refinement of AI-generated teaching videos from chatbot student interactions to create engaging educational content [56]. In fact, the work by [56] aimed to develop an AI generator capable of converting chatbot student dialogues into engaging educational videos to facilitate a more efficient student learning experience. Furthermore, AI algorithms should be able to understand and respond to human emotions, as well as social cues, in virtual contexts. These prove highly beneficial in educational settings because they enhance student engagement and interaction within the metaverse [106].

AI-based virtual assistants and chatbots play a pivotal role in enhancing interaction and engagement in the metaverse. Building on this, a significant challenge lies in developing virtual teaching assistants that are suitable for individual learners. This involves creating AI systems that can adapt to the unique learning styles and needs of each student, which requires sophisticated algorithms and a deep understanding of educational psychology [6]. In addition, AI must also facilitate seamless communication between learners from different linguistic backgrounds. This requires advanced natural language processing capabilities to ensure that language barriers do not hinder the educational experience. Achieving this level of language processing is a significant technical challenge [6,107]. Concerning practical application and real-world simulation, while AI can enhance interactivity, it cannot fully replicate real-world experiences. This limitation is particularly evident in fields that require hands-on practice, such as medical training, where AI-driven simulations may not capture all the nuances of real-world scenarios [108].

Tailored ML approaches are also necessary is in computer vision. Computer vision has the potential to significantly enhance the educational metaverse by enabling the analysis and recognition of human movements and actions. However, the specific environmental conditions of educational contexts, such as continuous interactions between teachers and students, necessitate the development of novel computer vision techniques and empirical investigations [54,75]. These techniques should be designed to support the unique conditions of the metaverse, ensuring that they can effectively facilitate learning and engagement [54]. Ref. [75] discusses the potential of computer vision techniques within the educational metaverse but does not specify particular techniques in detail. Instead, it emphasizes the need for adapting existing computer vision methods to the unique conditions of the metaverse. Another critical area is the development of neural interfaces that allow seamless access to the metaverse. Current neural interfaces, such as XR headsets, can cause physical discomfort when used for extended periods, which is a significant concern in educational settings, because the usage can be sustained over an extended period. Tailored ML approaches could focus on optimizing these interfaces to minimize negative physical effects, such as headaches and eye strain [54]. Furthermore, AI techniques should be developed to address the psychological and social aspects of the educational metaverse. For instance, emotion recognition algorithms could be tailored to identify signs of psychological stress or depression, providing timely alerts to educators. This would help in maintaining the mental well-being of students who might experience isolation due to the inherently distanced nature of the metaverse. These challenges are critical to address for the effective integration of AI technologies in educational settings within the metaverse.

4.4.2. Algorithmic Bias

In an educational context, algorithmic bias can impact everything from personalized learning paths to student performance assessments, which leads to the unfair treatment of students. In addition, based on characteristics such as race and socioeconomic status, bias can result in an unfair advantage for certain groups of students. For example, content or assessments may be more aligned with the experiences of certain cultural groups, to the detriment of others who do not share the same background [106,108]. This highlights the importance of the evaluation and the adjustment of AI algorithms to prevent discriminatory practices and ensure equitable educational opportunities.

On the other hand, biases can manifest in the accessibility of educational technologies. If the design and implementation of these technologies do not take into account the diverse needs of users, it can lead to inequitable access, especially for students with disabilities or those from underrepresented communities [69]. Furthermore, biased feedback can undermine students’ confidence and motivation, which may negatively impact their learning experience [106]. Therefore, establishing fair and accurate AI methods is important for the credibility of AI use in the metaverse for education [6,106].

4.4.3. Ethical and Security Concerns

The use of machine learning algorithms in educational contexts raises a number of ethical and security considerations [106,108]. In fact, the lack of specific regulations and the need for extensive data collection for personalized experiences and the training of AI models raise privacy concerns [69,107,108]. For example, to create a personalized learning environment, metaverse education systems need to collect identity information and personal data, such as learning preferences and emotional responses [6]. In preschools, the risk is higher because children may not be aware of the volume of data they are providing. Thus, maintaining privacy is often at the discretion of platform owners [107]. It is therefore essential to give importance to the rights of young learners to protect their data when integrating AI technologies into the metaverse in educational environments. The study by [6] affirmed the importance of ensuring that AI decision-making maintains trust and integrity by aligning with ethical standards, highlighting two specific security concerns within the metaverse educational environment: firstly, the misuse of AI, such as exploiting AI to cheat on exams or plagiarize work, and secondly, the issue of data security and privacy when utilizing real-time data to develop digital twins, which necessitates high precision and the ability to identify and rectify errors. The study by [6] asserted the importance of ensuring that AI decision-making preserves trust and integrity by aligning with ethical standards. In this context, the authors highlight two specific security concerns within the metaverse educational environment. First, the misuse of AI, such as exploiting AI to cheat on exams or plagiarize assignments. Second, the issue of data security and privacy when using real-time data to develop digital twins. The study by [108] argued that certain educational topics, such as ethics and safety compliance, are difficult to address effectively within the metaverse. Therefore, the use of AI systems must be carefully adopted to handle these sensitive topics, which adds another layer of complexity to their implementation. Furthermore, privacy and security issues tend to arise due to the interactions between chatbots and users that may include sensitive information [56]. So, the use of chatbots in the metaverse for educational purposes requires stringent measures to protect against breaches of security. On the other hand, the security concerns pose significant barriers to developing AI techniques used as an assistive technology for students with specific learning difficulties [69].

4.4.4. Integration and Interoperability Challenges

Integrating AI with the existing metaverse can be complex, requiring careful consideration of interoperability. Indeed, the metaverse relies on a wide range of technologies, which dictates the need for a unified and integrated system that combines virtual learning environments with metaverse technologies [6,108]. However, achieving this integration is complex due to the diverse range of technologies involved, including AI, blockchain, AR, VR, and MR. Each of these technologies has its own set of standards and protocols, which can complicate their integration into a cohesive educational platform. Ensuring that these technologies work together harmoniously is a significant challenge. Technical implementation challenges also contribute to interoperability issues. As an example, the AI-based generation process of a class notes by Ho et al. [61] involves various technologies, including Optical Character Recognition (OCR), Automatic Speech Recognition (ASR), and text-to-speech. Ensuring seamless integration and functionality among these technologies can be challenging and may require extensive testing and refinement. Concerning AI-based chatbots and virtual assistants, integrating their functionalities with existing metaverse infrastructure demands interoperability with various systems, requiring robust APIs and frameworks [56]. Further concern regarding interoperability is the lack of unified standards for integrating AI within the metaverse [56].

4.4.5. Computational Challenges

The use of AI in the metaverse can be limited due to the need of high computational processing and advanced hardware, which can be prohibitively expensive and not universally accessible [107]. On the one hand, the related computational demands require a robust infrastructure and significant processing power to handle AI algorithms, in particular, deep learning algorithms, along with real-time interactions of students [106] and 3D simulation content [107]. On the other hand, there is a need for extensive data transfer, which can be difficult to manage and necessitate the use of 5G or 6G networks and edge computing [107]. This may raise concerns about the digital divide, particularly as not all students and educational institutions may have the necessary technological resources.

Furthermore, the infrastructure layer of the metaverse, which includes AI, is still in its emerging phase. This infrastructure is not yet fully developed, which poses a challenge to the implementation of AI in the metaverse for education [108]. For example, to handle real-time simulations and advanced graphics, such as ray tracing, AI models must process vast amounts of data from IoT devices, digital twins, and geospatial sources to provide immersive visualizations in real time [13]. Balancing computational requirements with the need for interactive and seamless experiences is crucial for education, where AI-based systems must effectively engage learners. Furthermore, developing efficient algorithms and predictive models that reduce latency while ensuring an immersive experience also remains a major challenge.

5. Discussion

In this section, we analyze the findings reported above, highlighting key limitations. These limitations could serve as a foundation upon which to construct a research agenda to address gaps in the area of integrating AI into the metaverse to deliver education.

In the literature explored for this study, we observed the potential for the current rapid advances in AI techniques to play a crucial role in the creation of an AI-driven metaverse to transform education. We established that the AI-driven metaverse is already being introduced into several specific educational disciplines enumerated previously in Section 4 (RQ1). Furthermore, we discussed the potential education disciplines that may benefit from the use of AI techniques within the context of the metaverse. We identified a certain number of educational disciplines that have not yet benefited from this technology, but which have the potential to adapt and integrate existing AI tools into the metaverse. We also noted that even disciplines that are already under investigation would benefit from linking several novel aspects using AI techniques integrated into the metaverse. For example, the metaverse for medical education contains numerous sub-fields which could effectively incorporate AI tools to achieve more efficient healthcare learning. Regarding language learning, only English and Chinese were found to have been studied. The remaining languages would also benefit from the AI-based metaverse. We did, however, conclude that not all disciplines are suitable to integrate using such tools, since disciplines that require the need to master skills may necessitate a skill-based learning approach.

The research we reviewed primarily focuses on potential benefits and design concepts, rather than practical, operational systems. While much of the focus has been on the progress of the metaverse, existing efforts remain at the design or prototype stage, with implementations limited to 3D environments and specific tools, such as the class notes generator mentioned by [56]. Notably, despite the reviewed papers including the word “metaverse” in their abstracts and keywords, the majority reduce the metaverse to a 3D virtual environment, disregarding its more extensive elements, such as augmented reality, blockchain technology, and social engagement. Indeed, only a few papers described the metaverse as a whole ecosystem, e.g., [56]. Moreover, the absence of empirical testing and operational frameworks constrains the ability to assess enduring efficacy, scalability, and the educational impact of AI-based solutions within the metaverse to benefit education.

Although most articles discussed the importance of data privacy and security, they did not delve deeply into specific strategies or technologies to address related concerns. The requirement for stringent measures to protect sensitive information was acknowledged, although detailed solutions are not provided. In addition, while several papers discussed various applications for AI-based chatbots in the metaverse for education, they did not explicitly address the full range of challenges that might arise in different educational contexts. Our analysis revealed that there exist numerous technical, security, and integration-related challenges relating to the deployment of AI-based chatbots or virtual assistants.

Regarding educational use case applications, AI-based chatbots are among the applications most commonly explored for enhancing educational experiences. However, there is limited research regarding use of AI and user behavior within immersive learning environments. In traditional 2D learning settings, AI has been applied to assess students’ concentration and engagement levels through eye tracking and facial emotion analysis, e.g., [109,110,111,112]. With the evolution of head-mounted displays (HMDs) equipped with advanced eye, hand, and face tracking capabilities, AI could take on a pivotal role developing built-in assistive tools that enhance learning through behavior analysis. A key challenge here concerns training AI models to interpret human emotions and social cues accurately. Moreover, the diverse types of data collected by metaverse devices, especially from various sensors, present various challenges due to differing characteristics. Bypassing such challenges will be critical when creating adaptive learning environments able to respond effectively to students’ needs. On the other hand, integrating AI-powered solutions with emerging technologies such as digital twins, IoT, AR, and XR to provide adaptive learning is still being explored, with significant gaps present in terms of understanding long-term outcomes and scalability [69]. To address these challenges, additional theoretical and empirical research, as well as cross-disciplinary efforts, is needed to develop frameworks that support the seamless fusion of these technologies to enhance the learning experience.

The technical challenges of integrating AI into the metaverse that we have identified are specific to educational purposes and are related to the adaptation of AI techniques to the educational metaverse. The remaining challenges are common to the metaverse in general. However, other challenges are more specific to educational settings. Security challenges are more difficult to address because students, in particular those of preschool age, are not necessarily aware of the significance of the information they are exchanging. Additionally, concerning accessibility, not all students may be able to access the necessary hardware, such as XR HMD or high-performance computers, which could create disparities in learning opportunities. Moreover, implementing metaverse and AI technologies could be complex and may require significant technical expertise. This could pose challenges for educators unfamiliar with such technologies.

A recent advanced category of AI is represented by generative artificial intelligence (GAI) models, which produce new original content by learning from existing training data. This content can take various forms, including text, images, audio, complex designs, and more. The ability to generate new content will revolutionize numerous applications across different domains. The adoption of these models for generating content in the metaverse has sparked significant interest [113]. In the specific field of education within the metaverse, GAI can offer a variety of exciting possibilities to achieve more efficient learning experiences [114]. Some prominent use cases of GAI in education within the metaverse include the automation of text generation through tools like chabots (e.g., GPT) and virtual assistants [115]. These tools enable realistic, dynamic conversations in virtual educational environments and provide new ways of producing and presenting information. Furthermore, by analyzing student behaviors and attention, GAI could provide personalized virtual assistance [116]. This could later learn students’ preferences and needs to generate efficient recommendation systems for more flexible and efficient learning experiences. Language learning and translation are also promising applications of text generation. Such applications incur significant transformations for education within the metaverse. Additionally, GAI can be used to create customized courses, automate quiz generation, design laboratory experiments, develop simulations, and generate 3D objects and other educational resources. As a result, content creation [117] plays a crucial role in greatly enhancing student engagement and offers unique, flexible learning experiences alongside highly realistic communication, intelligent interactivity, and a dynamic virtual educational environment. As generative AI continues to evolve, it will open up new possibilities for transforming education within the metaverse. We expect that continued advances in artificial intelligence will unlock new promising opportunities, shape the future of education within the metaverse, drive innovation, and push the boundaries of creativity in this field. However, alongside the transformative power of GAI, it is important to understand and address the related challenges and barriers [118]. Among the most common issues, we cite the ethical considerations [119,120,121] such as intellectual property, data privacy, and user consent. Furthermore, the quality, realism, and authenticity of generated content should be handled responsibly. In fact, generated data are subject to misuse using tools such as deepfakes [122]. Addressing such issues can be achieved by implementing ethical, legal, and regulatory standards [123]. Another significant challenge is the high computational cost associated with training generative models. Addressing these issues and barriers is crucial to reduce the difficulty of obtaining high-quality training data, maintain trust, and encourage the widespread adoption of GAI approaches.

6. Recommendations and Future Directions

Our study has determined that, while AI has the potential to revolutionize education in the metaverse, it is imperative to address the challenges highlighted in Section 4 (RQ4) if its implementation is to be successful. In this section, we offer several recommendations to serve as potential directions for future research.

6.1. AI Algorithms

Combating algorithmic bias in the integration of AI into the metaverse necessitates several strategies. As solutions, we propose investing in Explainable AI (XAI). XAI is a set of procedures and techniques that allows users to understand and trust results [124]. It improves accountability by providing transparent information about AI decisions, identifying potential biases, and increasing confidence in the recommendations given. In addition, data diversification should be considered, as diversification can help minimize socioeconomic, cultural, and demographic biases. At the bias detection level, methods should be found to involve stakeholders, including teachers and students, to ensure biases are detected early in development. Moreover, implementing discrimination-aware unit tests at each stage of the algorithmic process and using open-source toolkits like Aequitas [125] and AI Fairness 360 [126] is recommended to monitor and mitigate potential biases. This ensures continuous evaluation and helps maintain fairness, impartiality, and effectiveness in educational technologies over time.

6.2. Security

Securely integrating AI into the metaverse is not a simple task and is one that requires a multifaceted approach. First, data anonymization plays a role in protecting user identities by concealing the link between personal information and the user. Second, strong encryption protocols and schemes can ensure secure data transmission. In this context, lightweight cryptographic schemes [127] represent promising candidates for use in environments such as the metaverse, where computing resources are critical. These schemes are designed to perform encryption and decryption efficiently, thereby reducing the burden on processing power, memory, and energy consumption. Third, blockchain integration helps to ensure secure and transparent data management in AI-powered virtual environments. Fourth, regular security audits, combined with continuous monitoring systems, can help detect vulnerabilities and mitigate threats in real time [107]. In addition, AI can play a pivotal role in addressing security challenges. The integration of AI with other technologies such as blockchain and non-fungible tokens (NFTs) offers a robust framework for enhancing identity security and managing digital assets. Moreover, AI can analyze vast amounts of data to authenticate users’ identities more effectively than traditional methods. This capability is particularly important in the metaverse, where digital interactions are frequent and require robust security measures to prevent identity theft and fraud. In this direction, distributed architectures supported by AI can facilitate secure interactions across 3D educational worlds, enhancing overall security [128]. Concerning authentication, AI algorithms can ensure seamless authentication by leveraging a multimodal approach that combines behavioral biometrics, such as typing patterns, voice recognition, gait analysis, etc. In the metaverse, particularly in educational settings, ensuring secure and seamless authentication is crucial for user experience and trust. Furthermore, developing AI algorithms to identify and counteract deep fakes is essential for maintaining the integrity of educational content and interactions in the metaverse [129]. Additionally, federated learning could usefully improve privacy by retaining user data on local devices.

6.3. Interoperability

Addressing interoperability challenges is crucial if it is going to be possible to deliver a seamless and effective educational experience in the metaverse. Among key strategies proposed is the establishment of universal interoperability standards across platforms to enable the seamless integration of AI applications into various virtual environments, thereby expanding access to educational content [108]. This universalization will seek to balance the interests of different stakeholders, including technology providers, educators, and learners. Some current technologies can be employed to address the interoperability challenges. For example, NFT technologies can be employed to monetize all metaverse components as unique assets, which allows for the seamless interaction and transfer of objects and assets between different metaverses [130]. This approach ensures that each digital asset, whether it is an avatar or any learning resource, is represented as an NFT that is unique. Furthermore, AI techniques can automate various processes involved in achieving interoperability, such as data conversion and protocol translation. The federated learning approach can also contribute to interoperability by ensuring that data remain on local devices. This method will not only protect privacy, but also facilitate the integration of various technologies by maintaining a consistent model across all devices [69].

6.4. Infrastructure

Maintaining a robust infrastructure is a key requirement when integrating AI into the metaverse seamlessly and efficiently, especially when the aim is to support real-time interaction between avatars in a 3D environment, which is important in an educational requirement. To transfer large data, in addition to the use of advanced communication networks, such as 5G and 6G, it is important to think about the efficient deployment and orchestration of computing and storage requests and resources. In this context, cloud services provide the computing resources necessary for the efficient operation of AI algorithms, even on devices that otherwise have limited processing power. Further research is needed into the collaboration mechanisms between edge devices and cloud systems in the context of the metaverse. For example, in this regard, Cloud-Edge-End Collaboration technology [131] could offer opportunities. In addition, the so-called green metaverse [132] must be prioritized by promoting approaches such as cooperative networking [133] and cooperative computation [134]. Cooperative networking helps to improve network efficiency by encouraging people to work together. For example, users who are located in close proximity to one another have common interests and so can share resources. Cooperative computing permits users to share computing resources. We believe that such approaches could provide encouraging and positive results in the context of the metaverse for education, with students, educators, and stakeholders expressing common goals and interests. In addition, several AI techniques can play a crucial role in addressing computational challenges. Deep reinforcement learning (DRL) can play a crucial role in efficiently allocating computing resources in the metaverse. On the one hand, DRL helps to optimize resource management in the dynamic environment [135]. On the other hand, DRL can be employed to adaptively balance loads in dynamic cloud environments, which is crucial for the metaverse’s fluctuating workloads [136]. Furthermore, federated learning allows for the decentralized training of models, reducing central computational loads. Another promising direction for optimizing resource allocation is the adoption of the UUT-centered AI paradigm, which is specifically designed for mobile edge computing in the metaverse and focuses on user and task-specific needs [137].

In Figure 9, we highlight the main research directions along with keywords representing specific objectives to achieve and some techniques to be employed.

7. Conclusions

This systematic review combines the understanding of AI and the metaverse as refined tools to further enhance the learning experience. The outcomes demonstrate that innovations such as those being developed have beneficial effects on the motivation, engagement, and participation of students in learning activities. In particular, such technologies enable the creation of AI-assisted metaverse environments that offer interactive, tailored, and adjustable learning, which will change the way people learn. With the help of these technologies, students are able to learn subjects that are intricate more effectively than any other way that has been in existence, resulting in an enhanced learning experience. This review also investigates those factors that might hinder integration of the metaverse and AI into the everyday practice of education. Even though the results were positive, there are challenges such as infrastructural difficulties, prejudice within algorithms, and privacy and safety issues, as well as moral questions. In light of these concerns, it is foremost that the emerging focus of scholarship is directed at strategies that are well defined that allow for the inclusivity and integration of AI and the metaverse into education in the future. For future practice, we have put forward a few practical ideas for researchers and educators. First, in addressing algorithmic bias, future research might seek a wider variety of AI training datasets in order to be more representative of the population and reduce biased results. Second, the linking of academic institutions and the business world would facilitate the growth of infrastructure, enabling schools and universities to use artificial intelligence and the metaverse.

Moreover, it is important to also develop behavioral and ethical recommendations for guidelines that make sure that necessary steps are taken, to ensure that students’ data are protected and that there is informed consent. For example, particular models such as decision transparency and auditing decision-making could help to safeguard sensitive information security concerns. Additionally, effort regarding research should not only bear in mind the general educational context but moreso focus on the consideration of particular educational structures such as primary and secondary and even higher education, thus allowing an understanding on the level of AI and metaverse application usability based on age and learning stage. In this scope of influencing factors, one can analyze the role of context in learning and artificial intelligence in some practical cases and guide particular settings. The implications for this are constructive, since one problem anticipates the existence of solutions tailored for the level of learners.

In conclusion, this literature review warns that if the impact of AI and the potential of the metaverse in education are not properly evaluated, the lack of solutions to global infrastructure challenges, ethical issues, and other technical problems will be further amplified. It is especially necessary to be well aware of issues when implementation strategies are designed and developed. Successful ongoing initiatives aiming to tackle these issues will ultimately dictate whether equitable access to the benefits of AI and teaching through the metaverse will become a possibility or not. As a result, all students will be assured of effective and equal participation in an emergent education seeking to exclude any barriers.