The Role of Artificial Intelligence in Gaming

Abstract

Artificial Intelligence (AI) has become a transformative force in the gaming industry, enhancing gameplay mechanics while expanding applications in education, healthcare, and human–computer interaction. The rapid growth of research in this domain requires a bibliometric review of its intellectual structure and thematic evolution. This study conducts a comprehensive bibliometric analysis of 5114 peer-reviewed documents indexed in Scopus and Web of Science between 2016 and 2025. Using performance indicators, co-authorship networks, and keyword co-occurrence mapping, the analysis identifies the most productive journals, countries, institutions, and authors contributing to the field. Citation analysis highlights seminal works on deep reinforcement learning as intellectual milestones, while keyword and thematic mapping reveal a dual trajectory: the development of advanced AI frameworks (reinforcement learning, deep learning, federated learning) and their application in societal domains such as education, psychology, and healthcare. The findings provide a consolidated overview of scientific production, intellectual influence, and thematic directions, contributing to a better understanding of the current state and prospects of AI–gaming research.

1. Introduction

Artificial Intelligence (AI) and digital games have coevolved for years. Games provide controlled, measurable, and diverse environments for testing AI ideas, while AI in turn enables richer, more adaptive, and creative gameplay experiences. Over the last decade, this relationship has accelerated with the rise of deep learning and, more recently, generative models, reshaping both fundamental research and industrial practices. A comprehensive review highlights how games serve as testbeds for AI methods such as planning, reinforcement learning, procedural content generation, and benchmarking, while simultaneously the unique requirements of game design stimulate methodological advances in AI [1]. In parallel, large language models have already been employed to design narrative content and quests, for example, role-playing games, demonstrating both their technical feasibility and the challenges of coherence, control, and safety [2]. At a more theoretical level, Capraro et al. have proposed a language-based game theory, illustrating how linguistic content influences strategic decisions, thereby moving beyond traditional game-theoretical frameworks centered only on economic payoffs [3].

The application of AI to player analytics and game balancing has also grown substantially. Large-scale telemetry allows the identification of emergent tactics and supports design decisions in real-world commercial settings [4]. Complementarily, behavioral profiling approaches have been developed to assess skill levels and playing styles, combining multimodal sensor signals with machine learning algorithms [5]. Predictive analytics further extends to churn detection, particularly in digital game-based learning (DGBL), where models such as logistic regression, decision trees, and random forests have been applied to predict dropout and retention rates, offering actionable insights for educational game designers [6].

Beyond gameplay mechanics, AI is increasingly shaping player interaction and learning through generative and adaptive approaches. For instance, the integration of ChatGPT into non-player characters within storyline-driven environments has been shown to affect students’ writing performance and engagement in game-based learning [7]. At the same time, the role of help-seeking behaviors when learners interact with AI-driven chatbots within DGBL scenarios is being studied, as these interactions can provide instrumental support without displacing essential cognitive processes [8].

Immersive technologies also interact with AI-driven gaming. Augmented reality games, such as AR-based alphabet learning systems for young children, have been evaluated for usability and acceptability, providing evidence of both pedagogical potential and practical challenges [9]. In addition, AI-based serious games are increasingly leveraged to address social issues. For example, video games that embed algorithms to detect gender bias in classroom interactions highlight how serious games can promote equity and challenge stereotypes in secondary education [10]. Similarly, AI frameworks have been developed to support socio-emotional skill development and emotion recognition in children with autism spectrum disorder, while serious game-based screening methods employing eye-tracking and deep neural networks offer innovative diagnostic support [11,12].

Within professional training and higher education, two converging trends can be observed. First, socially intelligent AI is increasingly integrated into cooperative game paradigms, where data-driven components of human–AI cooperation are investigated for their impact on engagement and shared agency [13]. Second, AI-driven virtual players have been embedded into simulation platforms for production management training, enabling adaptive difficulty levels and enhancing decision-making under realistic constraints [14]. In parallel, higher education is witnessing novel uses of generative AI for designing digital escape rooms, where students co-create game-based learning experiences, thereby developing creativity, problem-solving, and digital literacy skills [15].

Furthermore, AI is making inroads into entrepreneurial and business training contexts. Recent work demonstrates how AI-generated scenarios within serious games can be used to simulate technology start-up decision-making processes, exposing learners to uncertainty, iteration, and innovation practices that mirror real-world challenges [16]. This intersection of AI, gaming, and entrepreneurship underscores the broader potential of game-based learning beyond conventional education and entertainment.

These spectrum of applications illustrates how AI in gaming spans a continuum from foundational approaches to highly applied contexts, as summarized in Figure 1.

In addition to mapping research production, recent work highlights concrete applications of AI in gaming, including automated gameplay agents, adaptive difficulty systems, procedural content generation, personalized learning in educational games, affective computing for emotion-aware interaction, and diagnostic or therapeutic use of AI-assisted serious games.

Considering this multifaceted scenery, the exponential growth of research on AI in gaming calls for a structured and systematic assessment. A bibliometric approach offers such a lens, enabling the quantitative mapping of publication dynamics, influential outlets, collaborative patterns, and thematic structures. By analyzing peer-reviewed articles indexed in Scopus and Web of Science between 2016 and 2025, this study provides a comprehensive overview of the intellectual architecture of the field. Particular attention is given to emerging domains such as generative AI for narrative design, adaptive and immersive gameplay, socially intelligent systems, and diagnostic applications of serious games. To guide this exploration, the following research questions are addressed: RQ1: How has the scientific production on AI in gaming evolved over the last two decades in terms of volume, impact, and disciplinary scope?; RQ2: Which journals, institutions, countries, and authors constitute the intellectual and collaborative core of AI–gaming research?; RQ3: What are the most influential publications in the field, and how do they shape the technological and applied trajectories of AI in gaming?; RQ4: Which thematic clusters and research fronts emerge from keyword and co-occurrence analyses, and how do they reflect the multidisciplinary nature of the field?; RQ5: What future research directions can be identified from the thematic and strategic mapping of AI–gaming literature?

2. Materials and Methods

The study was designed to provide a comprehensive overview of AI in gaming through systematic bibliometric analysis. This methodological choice allows for the quantitative exploration of publication patterns, research networks, and thematic developments, thereby offering a reliable picture of the intellectual structure of the field [17,18]. To ensure transparency and reproducibility, the analysis was conducted in accordance with the PRISMA 2020 statement [19,20], which clearly documents the processes of identification, screening, and inclusion.

Two of the most authoritative and widely used bibliographic databases—Scopus and Web of Science (WoS)—were selected as data sources. These platforms are recognized for their high-quality indexing standards, broad coverage of peer-reviewed journals, and compatibility with Bibliometrix, the R-4.4.2 software package employed in this study [21]. The literature search was carried out in October 2025, covering a 10-year period (2015–2025). This timeframe was selected to focus on the most recent and significant advances in AI applied to gaming, particularly following the emergence of deep learning and reinforcement learning paradigms that revolutionized the field after 2015.

The construction of the search query was guided by two core conceptual dimensions, as illustrated in Figure 2. Search query dimensions for bibliometric analysis. The first dimension encompassed descriptors related to artificial intelligence, including a broad range of methodological paradigms and algorithmic approaches (“artificial intelligence” OR “machine learning” OR “deep learning” OR “Monte Carlo tree search” OR “reinforcement learning” OR “procedural content generation” OR “minimax” OR “alpha–beta prun*” OR “genetic algorithm*”). The second dimension captured the gaming domain, defined through complementary descriptors (“video game*” OR “digital game*” OR “serious game*” OR “computer game*” OR “game development” OR “game-based learning”). Records were included if at least one term related to artificial intelligence or machine learning appeared in the Title, Abstract, or Keywords. The two sets of terms were internally connected using the Boolean operator OR, which allowed the retrieval of documents containing any of the synonymous or closely related expressions within each conceptual dimension. To account for variations of word endings and plural forms, the truncation symbol * was also applied. The two dimensions—artificial intelligence and gaming—were then combined with the Boolean operator AND to ensure that retrieved records were positioned precisely at their intersection. The search was applied to titles, abstracts, and author keywords in both databases and restricted to documents written in English.

The process of document selection followed the PRISMA workflow, which is presented in Figure 3. The initial query retrieved a total of 7724 records (Scopus: 5350; WoS: 2374). These records encompassed different publication types, including journal articles, conference proceedings, reviews, book chapters, and other scholarly outputs. After removing 1854 duplicate records, the dataset was reduced to 5870 unique entries. A first screening was then performed on titles and abstracts. At this stage, documents falling outside the scope of the study were excluded. Specifically, reviews or conference review papers (385), books or book chapters (207), non-English publications (98), and errata, letters, retracted items, notes, or short surveys (64) were removed. Following this refinement process, a total of 5114 documents—comprising both journal articles and conference papers—were retained for analysis. These two publication types are particularly relevant in this field, as they represent the main channels through which original and peer-reviewed research on artificial intelligence in gaming is disseminated. Including both categories ensured a comprehensive and up-to-date overview of the research landscape, capturing not only established academic contributions but also the latest advances frequently presented in conferences.

3. Results

The final dataset obtained after applying the PRISMA protocol consisted of 5114 documents published between 2016 and 2025, as detailed in the previous section. Figure 4 summarizes the main descriptive statistics of the database. These records were distributed across 1702 different sources, reflecting the strong multidisciplinary nature of research on artificial intelligence and gaming. The analysis reveals an annual growth rate of 10.25%, indicating a steady and sustained increase in academic interest over the past decade. The average document age of 4.02 years confirms that this is a young and dynamic research area, with most contributions emerging in recent years. Despite this recency, the documents exhibit an average of 12.42 citations per publication, evidencing the rapid scholarly visibility and growing impact of the topic.

Regarding content descriptors, the dataset includes 10,044 author keywords and 14,854 Keywords Plus, demonstrating both the diversity of terminology used by researchers and the existence of complementary conceptual relationships identified through citation analysis. Together, these indicators illustrate the richness and evolving structure of the field, typical of areas undergoing intellectual consolidation.

In terms of authorship, the dataset comprises 13,327 contributing authors, among whom 389 are responsible for single-authored documents. The average number of co-authors per document is 3.86, while the international co-authorship rate stands at 9.66%. Here, a predominantly collaborative research culture, although collaboration tends to occur more frequently within institutions or national networks rather than across countries, consistent with patterns observed in emerging interdisciplinary domains such as AI in gaming.

Finally, regarding document types, the dataset is primarily composed of conference papers (n = 2266), journal articles (n = 1638), and proceedings papers (n = 1184), with a small number of early access articles (n = 26). The inclusion of both journal and conference contributions ensures a comprehensive and balanced representation of the field, capturing both well-established research outputs and cutting-edge developments typically presented at major conferences.

The evolution of annual publications reveals a consistent upward trend in research on AI and gaming over the last decade (Figure 5). The field experienced a strong acceleration starting in 2016, with an initial count of 106 publications, followed by a steady rise that peaked in 2024 with 829 documents. This sustained growth demonstrates the increasing academic and technological relevance of the field. The data indicate that between 2017 and 2023, the number of publications remained relatively stable, averaging around 550 records per year—reflecting a consolidation period in which the foundational concepts of AI in gaming matured and diversified. The sharp increase in 2024 corresponds to the recent wave of developments in generative AI and large language models, which have expanded research opportunities in adaptive gameplay, intelligent agents, and immersive virtual environments.

In parallel, the mean citation rate per article shows a declining trend, which is typical in rapidly growing research areas: as the volume of new publications increases, recent works have had less time to accumulate citations. Nevertheless, the continuous growth in output underlines the vibrant and evolving nature of the field.

It should be noted that the value for 2025 appears incomplete, as data collection was performed in October 2025, and thus does not cover the entire year. However, with 255 documents already indexed, it is expected that the total for 2025 will surpass previous records once the year concludes.

The 5114 documents analyzed were published in 1702 different sources, which evidences the multidisciplinary nature of research. Despite this broad dispersion, publication activity is concentrated within a limited number of core outlets, in accordance with Bradford’s Law of Scattering (

Table 1. Core scientific journals publishing on AI in gaming (top 10 by number of publications).

Source	Ranking	Frequency	Cumulative Frequency	Zone
Lecture Notes in Computer Science	1	250	1340	Zone 1
ACM International Conference Proceeding Series	2	129	1469	Zone 1
CEUR Workshop Proceedings	3	98	1567	Zone 1
IEEE Transactions on Games	4	86	1653	Zone 1
Communications in Computer and Information Science	5	75	1728	Zone 1
IEEE Access	6	74	1802	Zone 2
IEEE Conference on Computational Intelligence and Games (CIG)	7	50	1852	Zone 2
Proceedings of the European Conference on Games-Based Learning	8	42	1894	Zone 2
Advances in Intelligent Systems and Computing	9	40	1934	Zone 2
Applied Sciences—Basel	10	37	1971	Zone 2

). On the one hand, the Zone 1 journals and proceedings represent the most productive sources, accounting for most published works in the field. The Lecture Notes in Computer Science series leads with 250 publications, followed by the ACM International Conference Proceeding Series (129), CEUR Workshop Proceedings (98), and IEEE Transactions on Games (86). In contrast, Zone 2 includes sources that, while slightly less prolific, still play a significant role in shaping the field’s evolution. Examples include IEEE Access (74), Communications in Computer and Information Science (75), and the IEEE Conference on Computational Intelligence and Games (CIG) (50). Other influential venues in this zone—such as the Proceedings of the European Conference on Games-Based Learning. Together, the publications distributed across Zones 1 and 2 illustrate a balanced ecosystem between highly specialized computer science outlets and broader applied science platforms, confirming that the field has entered a stage of consolidation supported by both academic and professional communities.

While productivity identifies the most active publication venues, citation-based indicators such as the h-, g-, and m-index provide complementary information into their scientific influence (

Table 2. Highly impactful scientific journals ranked by number of publications, total citations, h-index, g-index and m-index.

Source	Number of Publications	Total Citations	h-Index	g-Index	m-Index	Publication Year Start
IEEE Transactions on Games	19	33	2.375	1321	86	2018
IEEE Internet of Things Journal	17	25	2.429	1593	25	2019
IEEE Access	16	33	2.000	1195	74	2018
Lecture Notes in Computer Science	16	27	1.778	1272	250	2017
ACM International Conference Proceeding Series	13	20	1.444	711	129	2017
Expert Systems with Applications	11	19	1.100	364	21	2016
IEEE Transactions on Computational Intelligence and AI in Games	10	16	1.000	290	19	2016
IEEE Transactions on Mobile Computing	10	15	1.250	365	15	2018
Procedia Computer Science	10	18	1.111	329	26	2017
Sensors	10	18	1.250	356	27	2018

). IEEE Transactions on Games emerges as both the most prolific and impactful outlet, with 19 publications, an h-index of 33, and a g-index of 1321, confirming its status as the central hub for disseminating cutting-edge research on AI in gaming since its first relevant publication in 2018. Other IEEE journals also demonstrate strong influence despite a smaller number of publications. For instance, the IEEE Internet of Things Journal (17 papers, h-index = 25) and IEEE Access (16 papers, h-index = 33) show how cross-domain journals have increasingly contributed to the dissemination of AI-driven game technologies, particularly in connectivity, embedded intelligence, and real-time computation. Lecture Notes in Computer Science (16 publications) and the ACM International Conference Proceeding Series (13 publications) remain essential venues for early-stage dissemination of research presented at leading international conferences. Their relatively high g- and m-indices highlight the cumulative relevance of conference-based contributions in shaping the field’s development.

Among specialized journals, Expert Systems with Applications and IEEE Transactions on Computational Intelligence and AI in Games (10 and 10 publications, respectively) demonstrate continued scholarly importance, bridging theoretical AI modeling with applied gaming contexts. In parallel, IEEE Transactions on Mobile Computing, Procedia Computer Science, and Sensors illustrate the increasing diversification of AI–gaming research toward topics such as mobile and pervasive computing, human–computer interaction, and sensor-based adaptive environments.

The analysis of author productivity according to Lotka’s Law (

Table 3. Author productivity through Lotka’s Law.

Documents Written	Number of Authors	Authors	Theoretical
1	10,762	0.808	0.618
2	1477	0.111	0.154
3	444	0.033	0.069
4	239	0.018	0.039
5	112	0.008	0.025
6	76	0.006	0.017
7	48	0.004	0.013
8	34	0.003	0.010
9	27	0.002	0.008
10	17	0.001	0.006

) reveals a highly skewed distribution, which is characteristic of emerging and interdisciplinary research areas. Most contributors are occasional authors, while only a minority exhibit sustained productivity over multiple publications. Specifically, 80.8% of authors produced a single publication, 11.1% contributed two papers, and less than 1% authored more than six. This pattern reflects a field that attracts a wide range of researchers from diverse disciplines—computer science, engineering, education, and psychology—who engage in AI–gaming research as part of broader scientific or technological efforts. When compared to the theoretical Lotka distribution, the observed pattern shows a notable deviation. The proportion of single-publication authors (0.808) is higher than the theoretical expectation (0.618), while the proportion of multi-publication authors is correspondingly lower. This imbalance suggests that the field is currently characterized more by breadth than depth, with extensive but fragmented participation and a relatively small group of core researchers maintaining long-term productivity.

Such trends are typical of research domains undergoing active expansion and conceptual diversification. The dominance of one-time contributors indicates that AI–gaming research is still consolidating its intellectual structure and institutional networks. As the field continues to mature, a gradual increase in multi-authored and recurrent contributions can be expected, leading to stronger collaboration patterns and greater thematic cohesion over time.

The country-level analysis (

Table 4. Country-level scientific production based on corresponding authors’ country (top 10 by number of publications).

Country	Articles	Articles %	SCP	MCP
China	676	13.2	607	69
United States	491	9.6	432	59
United Kingdom	202	3.9	159	43
India	178	3.5	171	7
Canada	154	3.0	119	35
Germany	154	3.0	123	31
Spain	146	2.9	125	21
Italy	136	2.7	119	17
Japan	122	2.4	107	15
Brazil	100	2.0	93	7

SCP = Single-Country Publications; MCP = Multi-Country Publications.

) distinguishes between single-country publications (SCP), which are authored exclusively by researchers from the same country, and multi-country publications (MCP), which involve international co-authorship and thus serve as indicators of cross-border collaboration. Here, research is geographically concentrated, with China (676; 13.2%) and the United States (491; 9.6%) leading global production. They are followed by the United Kingdom (202; 3.9%), India (178; 3.5%), Germany (154; 3.0%), and Spain (146; 2.9%), confirming the strong involvement of both Asian and European institutions. Collaboration patterns vary across countries. Canada (22.7%), Germany (20.1%), and the United Kingdom (21.3%) show higher international participation, whereas China (10.2%), India (3.9%), and Brazil (7.0%) display a more national orientation. Thus, the field exhibits a global but uneven structure, with North America and Europe acting as international hubs and Asia focusing on domestic research networks.

The institutional analysis (

Table 5. Leading institutions with the highest number of published articles (top 10 by number of publications).

Institution	Country	Published Articles
DeepMind Technologies Limited	United Kingdom	124
University of London	United Kingdom	59
Beijing University of Posts and Telecommunications	China	56
University of Alberta	Canada	55
University of California System	United States	48
Shanghai Jiao Tong University	China	46
Tsinghua University	China	46
New York University	United States	44
Bina Nusantara University	Indonesia	42
Shenyang Aerospace University	China	42

) reveals that research is driven by a combination of leading private laboratories and top universities. DeepMind Technologies (UK) stands out as the most productive institution with 124 publications, underscoring the critical role of industrial research in advancing AI breakthroughs applied to gaming. Academic institutions such as the University of London (UK, 59), Beijing University of Posts and Telecommunications (China, 56), and the University of Alberta (Canada, 55) also play major roles, followed by the University of California System (USA, 48) and Shanghai Jiao Tong University (China, 46). Chinese institutions dominate in volume, reflecting their rapid expansion in AI-related research, while North American and European universities contribute through well-established academic networks. The presence of both industrial leaders and academic centers underscores the complementary nature of innovation in this field, where collaborations between universities and private labs accelerate progress across technological, educational, and applied domains.

Figure 6 presents the geographical distribution of scientific production by country. The United States (1292 documents) and China (1141) dominate global output, followed by the United Kingdom (435), Canada (334), and India (317). In Europe, Germany (313), Spain (292), Italy (273), and France (184) stand out for their sustained productivity, while Japan (223) and South Korea (156) lead in Asia alongside China and India. Latin American contributions are led by Brazil (209), Mexico (56), and Chile (44), reflecting the gradual expansion of AI–gaming research beyond traditional technological hubs. The overall pattern indicates that North America, Asia, and Western Europe concentrate most global production. When related to institutional data (Table 5), this distribution reveals that national leadership is reinforced by key institutions—such as DeepMind Technologies in the UK, the University of Alberta in Canada, and several major Chinese universities—that act as focal points driving both national and international collaboration.

Table 6. Most cited documents, with total citations and annual impact (top 10 by number of total citations).

Ref	Authors	Scientific Journal	Total Citations	TC per Year	Normalized Total Citations
[22]	O. Vinyals et al.	Nature	3277	468.14	158.09
[23]	D. Silver et al.	Science	2830	353.75	26.04
[24]	S. Elfwing et al.	Neural Networks	1517	189.63	64.58
[25]	B. M. Lake et al.	Behavioral and Brain Sciences	1247	138.56	52.20
[26]	J. Schrittwieser et al.	Nature	962	160.33	60.38
[27]	M. Moravčík et al.	Science	745	82.78	12.71
[28]	A. Tampuu et al.	PLOS One	568	63.11	23.78
[29]	P. Li et al.	Future Generation Computer Systems	438	48.67	18.34
[30]	D. Holden et al.	ACM Transactions on Graphics	422	46.89	17.67
[31]	M. Jaderberg et al.	Science	380	54.,29	18.33

lists the most cited documents in AI–gaming research, illustrating the studies that have had the greatest scientific influence in shaping the field. The leading works include those by Vinyals et al. (3277 citations) [22] and Silver et al. (2830 citations) [23], both of which represent major milestones in deep reinforcement learning and self-play algorithms applied to complex games. These contributions established the foundations for current AI architectures used in decision-making and strategy learning. Other highly cited publications, such as Elfwing et al. [24], Lake et al. [25], and Schrittwieser et al. [26], further demonstrate the field’s rapid methodological evolution and interdisciplinary reach, connecting computational intelligence with neuroscience and behavioral modeling. Therefore, citation patterns confirm that deep reinforcement learning remains the central paradigm driving both publication growth and academic impact in AI–gaming research. The diversity of journals represented—from Nature and Science to PLOS One and Future Generation Computer Systems—also reflects the multidisciplinary dissemination of these advances across computer science, psychology, and applied engineering.

4. Discussion

4.1. Core Themes of High-Impact Studies

The most cited publications in the field of AI and gaming (Table 6) can be grouped into four thematic areas: (i) deep reinforcement learning (DRL) and general game playing, (ii) multi-agent and competitive environments, (iii) cognitive and behavioral modeling, and (iv) computational and algorithmic advances for game systems.

DRL and general game playing form the most influential cluster. The most influential works belong to this first cluster, which redefined the relationship between artificial intelligence and games. Silver et al. (2018) [23] presented AlphaZero, integrating deep neural networks with Monte Carlo Tree Search to achieve superhuman performance in chess, shogi, and Go through self-play, a fully autonomous learning approach that required no human heuristics. Building upon this foundation, Schrittwieser et al. (2020) [26] introduced MuZero, which unified model-free and model-based reinforcement learning by enabling the agent to learn internal models of the environment’s dynamics without direct access to game rules. These studies established games as benchmark environments for developing general-purpose AI systems capable of strategic reasoning and adaptive planning.

Multi-agent and competitive environments represent a second thematic cluster. A second thematic cluster centers on multi-agent reinforcement learning in complex and adversarial settings. Vinyals et al. (2019) [22] achieved grandmaster-level play in StarCraft II, overcoming challenges of imperfect information and massive decision spaces. Similarly, Moravčík et al. (2017) [27] introduced DeepStack, which reached professional-level performance in heads-up no-limit poker by combining recursive reasoning and deep neural networks for hidden-information domains. Jaderberg et al. (2019) [31] further advanced this line with population-based reinforcement learning, enabling adaptive cooperative and competitive behaviors in multi-player 3D environments. These contributions established games as ideal experimental frameworks for multi-agent and adversarial AI research.

Cognitive and behavioral modeling define a third area. The third cluster links AI development with human cognition and psychological learning processes. Lake et al. (2017) [25] proposed that artificial systems should learn and think like humans, emphasizing compositionality, causality, and intuitive reasoning. Elfwing et al. (2018) [24] complemented this perspective by introducing biologically inspired reinforcement mechanisms that emulate human-like adaptability and self-organization. These studies positioned games as testbeds for understanding and replicating human intelligence, bridging AI with cognitive science and behavioral psychology.

Computational and algorithmic advances for game systems constitute the fourth cluster. Finally, several studies highlight the development of algorithmic and computational methods aimed at improving learning efficiency, realism, and adaptability in game environments. Tampuu et al. (2017) [28] explored cooperative and competitive deep Q-learning in multi-agent contexts, while Li et al. (2017) [29] and Holden et al. (2017) [30] applied machine learning and motion synthesis techniques to enhance real-time decision-making and animation generation in interactive simulations. Algorithmic innovations provide the technical foundation enabling scalable and efficient AI behavior within complex virtual worlds.

In this regard, indicated high-impact studies outline two converging trajectories: on one hand, landmark advances in reinforcement learning that established gaming as a proving ground for general AI; and on the other, cross-disciplinary developments linking AI to human cognition, algorithmic optimization, and behavioral modeling. This dual evolution underscores the multidisciplinary and transformative nature of AI–gaming research, explaining its accelerated growth and global impact over the past decade.

4.2. Thematic Structure Based on Keywords and Co-Occurrence Analysis

The keyword analysis (Figure 7) provides a systematic overview of the most frequently used terms in AI–gaming research, highlighting both its conceptual foundations and methodological directions. The most frequent terms are computer games (1637) and artificial intelligence (1407), reflecting the dual focus of this domain on both technological development and application contexts. Core methodological keywords such as deep learning (826), machine learning (743), and reinforcement learning (701) emphasize the predominance of data-driven and adaptive models as the foundation of innovation in this field. Thematic diversity is evident through terms like human–computer interaction (546), serious games (452), and virtual reality (249), which reveal the growing integration of AI techniques in immersive, educational, and interactive environments. Keywords such as game theory (333), procedural content generation (253), and adversarial machine learning (209) further highlight the interplay between strategic decision-making, content automation, and robust AI behavior.

The co-occurrence network (Figure 8) complements this perspective by mapping the relationships between keywords, revealing five clearly differentiated clusters. The first cluster (blue) is centered on artificial intelligence and connects with serious games, deep learning, virtual reality, and gamification. It represents the technological axis of the field, where AI acts as the main driver of innovation in applied, educational, and training environments, fostering adaptive, immersive, and interactive learning systems. The second cluster (green) is anchored in machine learning, video games, and human, and is complemented by terms such as controlled study, classification, and physiology. This cluster emphasizes experimental and behavioral research, using video games as controlled settings to analyze human decision-making, cognitive performance, and physiological responses. The third cluster (red) revolves around game-based learning, including teaching, students, and adaptive learning. It reflects the educational and pedagogical dimension of AI–gaming, focused on the development of intelligent tutoring systems and the integration of AI-enhanced environments in formal education. The fourth cluster (purple) brings together terms such as reinforcement learning, adversarial machine learning, and learning algorithms, representing the algorithmic and computational core of the field. This group captures the advances in learning architectures, optimization, and robustness, which underpin progress in AI performance within gaming environments. Finally, the fifth cluster (yellow) links procedural content generation, game design, and virtual environments, highlighting the creative and design-oriented applications of AI. This area focuses on the automation of content creation, dynamic gameplay generation, and the development of personalized player experiences through machine learning techniques.

Beyond these clusters, the network reveals interconnections that reinforce the multidisciplinary structure of the field. For instance, links between reinforcement learning and serious games illustrate the transfer of algorithmic advances to educational contexts, while connections between virtual reality and game-based learning point to the integration of immersive and pedagogical paradigms. Similarly, the presence of adversarial machine learning across algorithmic and educational clusters underscores the growing attention to robustness, transparency, and ethical implications in AI applications for gaming.

The analysis of keyword frequency and co-occurrence reveals that AI-gaming research is evolving along two complementary axes. The first axis is the development of advanced technological frameworks, for example, reinforcement learning and deep learning which frequently appear as dominant AI techniques in this field [32]. The second axis is the application of these AI frameworks in societal domains such as education, psychology, and human–computer interaction. For instance, AI-driven methods are increasingly used to personalize and adapt educational games to individual learning needs [33], and in healthcare contexts AI-powered serious games are being applied for rehabilitation and cognitive training, yielding high user motivation and engagement [34]. This dual trajectory is consistent with the most cited works in the field, where methodological breakthroughs in AI align with applied studies in education, cognition, and healthcare.

This observed structure underscores the inherently multidisciplinary essence of AI–gaming research. The field serves as a fertile interface between computer science and human-centered disciplines, bridging technical innovation with behavioral sciences and educational innovation. Such a convergence ensures that progress in AI–gaming research drives both the sophistication of game AI technology and its meaningful application to societal challenges, solidifying the role of this domain as a nexus of technological and educational advancement.

In this regard, a comparative perspective reveals that reinforcement learning excels in autonomous decision-making in complex environments but requires large computational resources; deep learning provides high representational capacity yet suffers from limited interpretability; classical machine learning methods remain efficient for player modeling but struggle with scalability; and procedural content generation offers adaptability but requires careful human oversight to ensure coherence. Understanding these strengths and limitations helps contextualize methodological choices across different game-related applications.

4.3. Thematic Evolution Based on the Strategic Diagram

The strategic diagram (Figure 9) provides a strategic visualization of the conceptual structure of research by plotting clusters along two axes: centrality, which represents the degree of relevance or connectivity of a theme within the field, and density, which indicates the level of internal development of that theme. Four quadrants emerge from this representation (motor themes, basic themes, niche themes, and emerging or declining themes) allowing for a nuanced interpretation of the field’s intellectual organization.

The first cluster, composed of machine learning, video games, and human, is located between the motor themes and basic themes quadrants. This positioning indicates that these topics are both central and relatively well developed, functioning as methodological and conceptual drivers of innovation. The trend identified here shows how video games increasingly serve as controlled environments for studying cognitive and behavioral patterns using machine learning. This trajectory has strengthened in recent years, driven by works applying physiological sensing, predictive analytics, and classification models to understand player behavior and human–AI interaction.

The second cluster, formed by artificial intelligence, deep learning, and computer games, lies firmly within the basic themes quadrant. Its high centrality and moderate density confirm that these concepts represent the foundational backbone of the domain, defining the technological base from which most other research areas emerge. The trend reflected here is the consolidation of a stable methodological base: deep reinforcement learning, neural planning models, and multimodal AI approaches that feed into both experimental and applied research.

The third cluster, including reinforcement learning and game theory, is positioned in the motor themes quadrant, denoting areas of high relevance and strong development. These topics drive technical and algorithmic progress, particularly in the creation of adaptive agents, strategic gameplay, and self-learning systems that push the boundaries of artificial intelligence in interactive environments. This trend is strongly supported by high-impact publications such as AlphaZero, MuZero, and multi-agent RL systems, which continue to push the boundaries of autonomous decision-making in complex environments.

Finally, the fourth cluster, encompassing serious games and game-based learning, occupies the emerging or declining themes quadrant. Although these topics have lower density, their notable centrality suggests that they are actively evolving as bridges between AI technologies and educational or training applications. The trend suggests a transition from early exploratory studies toward more sophisticated, evidence-based frameworks integrating AI with educational design and learning sciences.

4.4. Gaming Types and Their Relevance in AI Research

To further contextualize the scope of AI–gaming research,

Table 7. Distribution of gaming types in AI–gaming research. * Term variations and improvement of the bibliometric search.

Gaming Type	Description	Keywords for Query Extension	Results
Video Games	Entertainment-focused interactive systems; used as testbeds for AI behavior modeling and gameplay.	“video game*”, “digital game*”, “computer game*”, “interactive entertainment”	65.96%
Serious Games	Games designed for education, training, or awareness rather than entertainment.	“serious game*”, “educational game*”, “training game*”, “simulation-based learning”	14.01%
Game-Based Learning	Integration of game elements into instructional design to enhance learning engagement and outcomes.	“game-based learning”, “gamification”, “educational technology”, “learning analytics”	9.09%
Simulation Games	Virtual environments modeling real-world processes (e.g., flight, surgery, or driving simulators).	“simulation game*”, “virtual training”, “serious simulation”, “immersive simulation”	0.64%
Augmented and Virtual Reality Games	Immersive, sensor-driven environments for entertainment, education, or rehabilitation.	“virtual reality”, “augmented reality”, “mixed reality”, “immersive game*”	9.35%
Health and Rehabilitation Games	Games used for therapy, physical rehabilitation, or cognitive training.	“rehabilitation game*”, “health game*”, “cognitive training”, “exergame*”	0.79%
Serious Simulations and Training Systems	Applied gaming systems used in defense, medicine, or industry.	“training simulator”, “military simulation”, “medical simulation”, “industrial training”	0.16%

categorizes the main gaming types identified in the literature, describing their primary functions, associated query keywords, and relative representation within the dataset. This classification reveals how different gaming modalities serve as experimental testbeds for various AI paradigms—ranging from algorithmic innovation in entertainment-oriented video games to adaptive learning environments and health-related applications.

The data shows that entertainment-oriented video games dominate the field, accounting for nearly two-thirds of all publications, which confirms their role as primary platforms for AI experimentation and benchmarking. In contrast, serious and educational games collectively represent around one-fourth of the research output, evidencing a strong and growing interest in pedagogical and training applications. Meanwhile, augmented reality and virtual reality and health-related games are emerging as cross-disciplinary extensions, connecting technological development with human-centered domains such as education, rehabilitation, and immersive learning. This distribution underscores the progressive diversification of AI–gaming research, moving from entertainment-focused studies toward broader societal and applied contexts.

4.5. Future Prospects

Looking forward, the evolution of AI in gaming opens promising but also challenging avenues. Ethical considerations will be central, as adaptive systems increasingly rely on the collection and analysis of sensitive data about players’ behavior, motivation, and learning performance. Issues such as privacy protection, transparency in algorithmic decision-making, and the prevention of manipulative practices are critical to ensuring that AI-driven game-based applications foster trust and responsible use in educational settings. Furthermore, the design of serious games should align with principles of inclusivity and fairness, avoiding the reinforcement of stereotypes and ensuring equal access for learners with diverse backgrounds and abilities.

At the pedagogical level, recent research emphasizes that the effectiveness of gamification and game-based learning depends on technological sophistication and the recognition of diverse player profiles [35,36]. Future studies should therefore focus on tailoring game mechanics and adaptive AI systems to different cognitive styles, motivational drivers, and cultural contexts, thereby maximizing engagement and learning outcomes in higher education. This perspective highlights the need to integrate insights from psychology, education, and human–computer interaction into the design of AI-enhanced games.

Another crucial avenue is interdisciplinarity. By combining technical advances in machine learning and generative AI with educational sciences, ethics, and social sciences, researchers can develop holistic frameworks that balance innovation with human-centered values. In parallel, sustainability should not be overlooked: the energy demands of large-scale AI models raise questions about the environmental footprint of AI-driven gaming ecosystems, suggesting that future research must also address green and efficient computational strategies.

These prospects point toward a balanced trajectory in which technological innovation is accompanied by ethical responsibility, pedagogical inclusivity, and sustainable practices. Such an integrated vision will ensure that AI in gaming continues to evolve as a powerful tool for education, training, and societal benefit.

5. Conclusions

This study provided a comprehensive bibliometric analysis of 5114 documents on Artificial Intelligence (AI) in gaming, indexed in Scopus and Web of Science between 2016 and 2025. Guided by the proposed research questions, the following conclusions can be drawn:

(RQ1) Scientific production on AI in gaming has shown sustained exponential growth over the last decade, driven by advances in deep learning and reinforcement learning. A marked citation peak corresponds to breakthrough contributions in deep reinforcement learning and self-play, which positioned gaming as a testing ground for general-purpose AI. The field has transitioned from a niche area into a mature and multidisciplinary domain, spanning applications in education, healthcare, psychology, and human–computer interaction.

(RQ2) The distribution of research across journals, institutions, and countries reveals a concentration of output in leading publication venues and highly active organizations. Lecture Notes in Computer Science, IEEE Transactions on Games, and IEEE Access dominate scientific dissemination, while DeepMind Technologies, the University of Alberta, and Beijing University of Posts and Telecommunications lead institutional productivity. Geographically, China, the United States, and the United Kingdom are the most prolific contributors, together accounting for more than one-third of all publications. North American and European institutions exhibit strong international collaboration networks, while Asian institutions display significant productivity within more nationally focused frameworks.

(RQ3) Citation-based indicators confirm the disproportionate influence of a small group of landmark publications, particularly those introducing deep reinforcement learning (e.g., AlphaGo, AlphaZero, MuZero) and multi-agent learning. These works transformed video games into experimental environments for algorithmic innovation and inspired applied studies linking AI to cognitive modeling, learning systems, and adaptive gameplay. The balance between foundational theoretical works and highly cited applied research underscores the dual scientific and technological trajectory of the field.

(RQ4) The analyses identified five thematic clusters in AI–gaming research: technological frameworks, human–computer interaction, educational applications, computational advances, and creative design. The strategic map positioned reinforcement learning and game theory as motor themes, AI and deep learning as core foundations, and serious and game-based learning as emerging areas, reflecting a dynamic balance between technological progress and human-centered innovation.

(RQ5) Looking ahead, the thematic evolution suggests multiple avenues for future research. On the technological front, opportunities lie in hybrid and federated learning systems, explainable AI, and robust reinforcement learning for adaptive and ethical gaming environments. On the applied side, promising directions include the integration of AI with immersive and affective technologies, the design of socially intelligent and emotionally aware agents, and the expansion of AI-driven serious games for education, therapy, and rehabilitation. Strengthening international collaboration, as well as connecting algorithmic innovation with human-centered applications, will be essential to sustain long-term growth and impact.

Future research is expected to advance toward explainable and transparent AI for improving trust in game-based systems, hybrid learning architectures that combine symbolic and neural approaches, and human-centered adaptive engines capable of modeling affect, motivation, and long-term learning behaviors. Ethical considerations (privacy, fairness, and responsible data use) will become increasingly important as AI-driven gaming applications expand into education, healthcare, and professional training contexts.

Finally, the results confirm that AI in gaming has developed along a dual trajectory: advancing cutting-edge AI frameworks while extending their applications to domains with direct societal impact. This duality underscores the multidisciplinary character of the field, positioning it at the intersection of computer science, psychology, education, and healthcare. By mapping its intellectual structure and research fronts, this study contributes to a deeper understanding of the field’s current state and provides a foundation for guiding future investigations and innovations in AI-driven gaming.