Global insights and the impact of generative AI-ChatGPT on multidisciplinary: a systematic review and bibliometric analysis

ABSTRACT

In 2022, OpenAI's unveiling of generative AI Large Language Models (LLMs)- ChatGPT, heralded a significant leap forward in human-machine interaction through cutting-edge AI technologies. With its surging popularity, scholars across various fields have begun to delve into the myriad applications of ChatGPT. While existing literature reviews on LLMs like ChatGPT are available, there is a notable absence of systematic literature reviews (SLRs) and bibliometric analyses assessing the research's multidisciplinary and geographical breadth. This study aims to bridge this gap by synthesising and evaluating how ChatGPT has been integrated into diverse research areas, focussing on its scope and the geographical distribution of studies. Through a systematic review of scholarly articles, we chart the global utilisation of ChatGPT across various scientific domains, exploring its contribution to advancing research paradigms and its adoption trends among different disciplines. Our findings reveal a widespread endorsement of ChatGPT across multiple fields, with significant implementations in healthcare (38.6%), computer science/IT (18.6%), and education/research (17.3%). Moreover, our demographic analysis underscores ChatGPT's global reach and accessibility, indicating participation from 80 unique countries in ChatGPT-related research, with the most frequent countries keyword occurrence, USA (719), China (181), and India (157) leading in contributions. Additionally, our study highlights the leading roles of institutions such as King Saud University, the All India Institute of Medical Sciences, and Taipei Medical University in pioneering ChatGPT research in our dataset. This research not only sheds light on the vast opportunities and challenges posed by ChatGPT in scholarly pursuits but also acts as a pivotal resource for future inquiries. It emphasises that the generative AI (LLM) role is revolutionising every field. The insights provided in this paper are particularly valuable for academics, researchers, and practitioners across various disciplines, as well as policymakers looking to grasp the extensive reach and impact of generative AI technologies like ChatGPT in the global research community.

KEYWORDS:

• LLM-ChatGPT
• bibliometric analysis
• interdisciplinary research
• demographic analysis
• generative AI
• geographic distribution

PACS:

• 0000
• 1111

2000 MSC:

• 0000
• 1111

1. Introduction

Artificial intelligence (AI) has witnessed a significant transformation in recent years, leading to notable advancements in chatbot technology. Among these advancements, ChatGPT has emerged as a highly significant AI language model that has demonstrated its potential to revolutionise conventional practices across diverse fields (Baidoo-Anu & Ansah, 2023).

ChatGPT is a powerful AI language model that has a vast pre-training corpus comprising 1.5 billion parameters, making it one of the largest language models in existence. The model is based on a transformer-based neural network architecture that utilises the attention mechanism to understand the context of the input text and generate coherent responses. Its architecture allows it to maintain consistency in tone and style, making it a highly effective tool for various natural language processing tasks. The success of ChatGPT in natural language processing and conversational AI can be attributed to its innovative architecture, which enables it to generate high-quality responses for various tasks, including question answering, language translation, and text summarisation. The model has been pre-trained on a massive amount of data, including text from the internet, books, and articles, which has enabled it to acquire a vast knowledge of language (Ray, 2023).

ChatGPT has shown tremendous potential in transforming traditional practices across various fields, including customer service, education, and healthcare. Its ability to generate human-like responses has made it an increasingly popular tool for chatbots and virtual assistants. The model's effectiveness in understanding the context and generating coherent responses has garnered widespread attention in AI, making it a significant milestone in developing conversational AI and natural language processing (PCMag, n.d.). Developed by OpenAI, ChatGPT was initially launched in November 2022 based on the GPT-3.5 architecture, followed by a subsequent release in March 2023 based on GPT-4.0, indicating the rapid evolution and adoption of this technology (R. Goodman et al., 2023a; OpenAI, n.d.a, n.d.b).

1.1. Objectives and scope

This study aims to delve into the interdisciplinary and demographic research bases of ChatGPT, exploring its application across various disciplines and examining the geographic distribution of ChatGPT-related publications. The objectives extend to providing a structured overview of the existing ChatGPT research landscape, identifying key institutions and authors, and shedding light on the global interest and inclusive accessibility of ChatGPT across different regions.

The scope of our analysis encompasses a comprehensive review of the literature and a bibliometric analysis to elucidate the interdisciplinary nature of ChatGPT applications and its demographic research base. This investigation covers a wide array of disciplines, including education, nursing, medical research, and various industrial sectors (Badshah et al., 2023; Gabashvili, 2023; Miao & Ahn, 2023; Ray, 2023). Our goal is to showcase the transformative potential of AI ChatGPT by integrating it across various disciplines and countries, where participants have been involved in the ChatGPT research. The research was limited as only two databases (Scopus and Web of Science) were used and other publisher databases were not examined.

1.2. Motivation and contributions

While several authors have conducted surveys on the applications of ChatGPT, these investigations have often been confined to specific disciplines. This limitation underscores the need for a more comprehensive, interdisciplinary approach to understand the full spectrum of ChatGPT's impact across various fields. This study addresses this gap by offering an extensive review and analysis that transcends disciplinary boundaries, providing a more holistic understanding of ChatGPT's applications and implications (Miao & Ahn, 2023; Ray, 2023; Sallam, 2023; P. Zhang & Tur, 2023). The key contributions of this study are outlined as follows:

• Providing a comprehensive, interdisciplinary review of ChatGPT applications, encompassing a wide range of academic and industrial sectors.

• Analysing the geographic distribution of ChatGPT-related research, highlighting its global impact and accessibility.

• Identifying leading institutions and prolific authors in ChatGPT research, mapping the academic landscape of this emerging field.

• Offering insights into the potential challenges and opportunities posed by ChatGPT in various disciplines.

The paper is structured into several key sections: Section 2 discusses related literature, while Section 3 analyses ChatGPT's applications across various fields. Section 4 examines the evolution and key features of ChatGPT. In Section 5, we explore its academic applications across disciplines. Section 6 addresses the challenges and opportunities of using ChatGPT. The paper concludes with a summary of our findings and reflections on future directions.

2. Related work

A thorough review of current literature reveals a growing interest in applying ChatGPT across various disciplines, notably in healthcare, customer service, and education (Qureshi, 2023). Ray (2023) conducted a systematic literature review (SLR) to provide a comprehensive overview of ChatGPT, covering its development and applications, while also addressing significant ethical concerns and safety issues. Notably, their study, while extensive, did not delve into the specifics of GPT architectures nor did it perform a comparative analysis between different versions of ChatGPT, such as contrasting ChatGPT 3.5 with ChatGPT 4. Similarly, Miao and Ahn (2023) investigate the impact of ChatGPT on nursing education, underscoring its potential to transform traditional teaching methods. Their review further explores ChatGPT's role in scientific research, from data analysis to enhancing public engagement in scientific endeavours (Roumeliotis & Tselikas, 2023). It acknowledges the ethical challenges and operational hurdles in research settings, emphasising the need for a balance between AI advancements and human expertise. Despite controversies, ChatGPT's rapid acclaim in academia and various industries, as highlighted in recent studies (Mhlanga, 2023), underscores its emerging status as a pivotal AI tool.

In our series of articles on ChatGPT, we have explored two key studies (Koubaa, 2023; Koubaa et al., 2023), with the first providing a detailed analysis of the GPT architecture and its application across various fields, and the second offering a comparative study of ChatGPT 3.5 and GPT-4, highlighting the evolutionary advancements in the series. These studies not only shed light on the technical nuances and capabilities of ChatGPT but also revealed its significant impact on AI. However, they did not extend to a systematic literature review or bibliometric analysis to examine ChatGPT's broader disciplinary and geographical influence. Given ChatGPT's growing attention from diverse sectors, our future work aims to address this gap by conducting an extensive analysis, thus contributing to a deeper understanding of its global reach and potential applications.

In their study (P. Zhang & Tur, 2023), the authors carried out a systematic literature review within the educational sector to explore the application of GPT architectures, specifically focussing on ChatGPT's role in Kindergarten to 12th-grade (K-12) educational settings. Adhering to the PRISMA framework, their research synthesises findings from 13 papers, offering a comprehensive SWOT (Strengths, Weaknesses, Opportunities, Threats) analysis of ChatGPT's implementation in K-12 education. The review highlights ChatGPT's potential to empower educators in curriculum development, lesson planning, materials generation, and enhancing personalised student learning experiences. It also addresses concerns about academic integrity and output quality, providing practical pedagogical recommendations and ethical considerations for the effective integration of ChatGPT in K-12 settings, particularly emphasising the role of collaboration and co-design in learning processes. While this study offers valuable insights into the use of ChatGPT in education, its focus is limited to K-12 education.

In their systematic literature review, Sallam (2023) focussed on ChatGPT's applications in healthcare, following PRISMA guidelines to analyse records from PubMed/MEDLINE and Google Scholar. The study, encompassing 60 records, highlighted ChatGPT's benefits in healthcare education, research, and practice, such as enhancing scientific writing, research efficiency, and health literacy, while also improving personalised learning. However, it raised concerns about ethical, legal, and accuracy issues, emphasising the need for a responsible approach to ChatGPT's adoption in healthcare. Notably, while providing substantial insights, the study did not compare different ChatGPT versions, indicating a gap in the current research. This calls for further exploration into the specific capabilities and evolution of ChatGPT's various iterations in the healthcare sector.

In the landscape of existing research on ChatGPT, as summarised in our comprehensive Table 1, our work distinguishes itself by adopting a holistic and multifaceted approach. Unlike the focussed studies such as Koubaa et al. (2023) which delved into the GPT architecture, and P. Zhang and Tur (2023) that concentrated on the use of ChatGPT in K-12 education, our survey transcends these specific domains. We not only explore the intricacies of the GPT architecture but also provide a comparative analysis between ChatGPT 3.5 and 4, a facet not extensively addressed in previous studies. Our work also stands out by incorporating an SLR and bibliometric analysis, offering a panoramic view of the utilisation of GPT architectures across multiple disciplines. This comprehensive scope is further enriched by extending our analysis to include the geographical distribution of research, thus providing a global overview of GPT advancements and applications. As detailed in Table 1, while other studies like Sallam (2023) focussed on healthcare applications and did not perform a version comparison, our work fills these gaps. Our contributions include an interdisciplinary review of ChatGPT applications across academic and industrial sectors, an analysis of the geographic distribution highlighting global impact and accessibility, identification of leading institutions and authors, insights into potential challenges and opportunities, and proposing future directions for ChatGPT research, emphasising the need for continued exploration and innovation. This multifaceted approach underscores our study's unique place in the ever-evolving research landscape of AI and ChatGPT.

Table 1. Comparison with existing work.

Papers	GPT Architecture	ChatGPT 3.5 vs 4 Comparison	SLR + Bibliometric Analysis	Multi-Discipline-wise Case Studies	Geographical Distribution
Koubaa et al. (2023)	Yes	No	No	No	No
Koubaa (2023)	No	yes	No	No	No
P. Zhang and Tur (2023)	No	No	SLR	Education	No
Miao and Ahn (2023)	No	No	No	Healthcare	No
Ray (2023)	Yes	No	No	No	No
Sallam (2023)	No	No	SLR	Health care	No
Our work	yes	yes	yes	yes	yes

3. Research methodology

This section outlines the wide range of ChatGPT applications across different fields. As a leading example of advanced AI, ChatGPT has significantly influenced various sectors, changing how knowledge is shared, learned, and developed. Our analysis is based on a detailed review of existing literature and a thorough bibliometric analysis, aiming to reveal the diverse interdisciplinary uses and global reach of ChatGPT in research.

The methodology employs a two-pronged bibliometric analysis and a systematic literature review.

3.1. Bibliometric analysis

The bibliometric analysis entailed a detailed examination of publications related to ChatGPT across various databases, including Scopus and Web of Science. Metrics such as the number of publications, citation indices, and the geographic distribution of authors were analysed to discern the research trends and the global impact of ChatGPT (Khosravi et al., 2023). The geographic data of the authors were collected from the Web of Science database for detailed analysis.

3.2. Systematic literature review

The PRISMA guidelines were strictly adhered to during the systematic literature review, which involved a rigorous search strategy to identify relevant studies. Databases like PubMed and Scopus Digital Library were thoroughly searched, and studies were chosen based on specific criteria for inclusion and exclusion. The primary objective of the review was to uncover the various applications of ChatGPT across different fields and to gain insights into the existing challenges and opportunities.

3.2.1. Data analysis

Qualitative and quantitative analyses were conducted on the data extracted from the chosen studies. The recurring themes and trends were identified through the utilisation of thematic analysis. Statistical analysis, on the other hand, was employed to make inferences and compare the influence of ChatGPT across various areas, indicating the fields that have extensively integrated research with AI.

3.2.2. Research questions

Although an interest in a particular sector is typically the catalyst for a research operation, having an understanding of the topic aids in developing a good research question. Questions arise from a perceived knowledge gap in a field or study area.

The research questions have been formalised for this survey to analyse different link and node failures and their recovery approaches. We have prepared The following research questions in Table 2 and hope to address them through this SLR.

Table 2. Research questions and motivation for the SLR.

Research Questions	Motivation
How have the capabilities and functionalities of ChatGPT evolved across its successive versions?	This inquiry aims to delineate the trajectory of ChatGPT's development, enabling a critical assessment of its performance enhancements and refined features over time.
In what manner has ChatGPT been employed within diverse academic and research domains?	The motivation here is to catalog and synthesise the broad range of ChatGPT applications, thereby illustrating its interdisciplinary utility and impact.
What is the global distribution of scholarly activity concerning ChatGPT, and which regions are at the forefront?	The objective is to map the global research landscape, highlighting regional contributions to the field, and discerning potential academic and cultural influences on ChatGPT's research progression.

3.2.3. Materials and methodology

• Search Databases and combination of keywords

In academic databases like Table 3 and the search keywords used in Table 4 in this SLR, Boolean expressions link keywords for data extraction. By combining terms, the Boolean Operators expand or focus the search queries by employing the terms AND and OR. The Booleans used in this SLR are shown in Table 5.

Table 3. Database sources search result and keywords.

DB-1	Scopus	Abstract, Title, keywords
DB-2	Web of science	Abstract, Title, keywords

Table 4. Search strings.

S.No	Batch 1=B1	Batch 2=B2
1	Chatgpt	Gpt
2		Large language model
3		LLM
4		Large language model(LLMs)
5		GPT-3
6		GPT-4

Table 5. Combination of strings.

S.No	Combination of Strings
1	(1, B1) AND (1, B2)
2	(1, B1) AND (2, B2)
3	(1, B1) AND (3, B2)
4	(1, B1) AND (4, B2)
5	(1, B1) AND (5, B2)
6	(1, B1) AND (6, B2)

• Inclusive and exclusive

The inclusion and exclusion criteria determine the systematic literature review's scope. They are primarily predetermined after deciding on the research issue and before searching, but scoping searches could be necessary to choose pertinent inclusion/exclusion criteria. Various aspects may be used to define these criteria as inclusion or exclusion criteria, as shown in Table 6. We have formulated a series of eligible studies since it is crucial for any systematic review to ensure that only results relevant to the study question are considered. Figure 2 shows that we have got analysis 88% of the journal papers and 12% of conference papers in this study according to our criteria, which we have selected above. In this study, we employed the PRISMA methodology to screen the collected data step by step and perform data analysis on the remaining data. The information is illustrated in Figure 1. Furthermore, we got information from this data of top 10 publisher as illustrated in Figure 3

Figure 1. Data collection through PRISMA methodology in this study (2022–2024).

Figure 2. Distribution of selected papers across journals and conferences based on our chosen dataset.

Figure 3. Top 10 publishers in the field of ChatGPT research along with their total citations.

Table 6. Inclusive and exclusive criteria for research articles scrutiny.

Inclusive criteria	Exclusive criteria
studies, which include ChatGPT and its application in multiple disciplines and studies that	involve other problems except ChatGPT
The study focuses on implication of chatgpt.	Studies that involve other problems
Articles written in English are considered.	Articles written in other languages are not included.
The scholarly and peer-reviewed journal articles should be considered. Journals with a good reputation or conference papers should be considered.	Articles that have not been published blogs, news are also not included.

4. Evolution of ChatGPT: key features and developments

RQ1: How have the capabilities and functionalities of ChatGPT evolved across its successive versions?

4.1. Overview of ChatGPT

ChatGPT has emerged as a significant innovation in NLP, a branch of AI focussed on enabling machines to understand and generate human language (Eysenbach, 2023; Fijačko et al., 2023; Haleem et al., 2022; Michail et al., 2023). The development of ChatGPT was driven by the goal of creating an AI language model that is versatile and sophisticated enough to handle tasks like text generation, translation, and data analysis.

The inception of ChatGPT is anchored in the advancements of the Transformer architecture (Vaswani et al., 2017), as shown in Figure 4. This development represented a major shift in the NLP landscape, addressing limitations of earlier sequence-to-sequence models such as Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs). The Transformer architecture was pivotal in developing OpenAI's GPT series, including GPT-2 and GPT-3, which laid the foundation for ChatGPT (Vaswani et al., 2017).

Figure 4. The transformers architecture as defined in Vaswani et al. (2017) for Machine Translation.

The Transformer architecture, as delineated in Vaswani et al. (2017) for Machine Translation, marks a significant advancement in natural language processing. Central to this architecture is its innovative use of self-attention mechanisms, which allow the model to weigh the importance of different words within a sentence, irrespective of their sequential position. This capability differentiates it from traditional models like RNNs and LSTMs that process data sequentially. By concurrently processing entire sequences of text, the Transformer enables more parallelisation, leading to substantial improvements in translation quality and efficiency. Its architecture, comprising multi-head attention and feed-forward networks, has become a foundational model in various language understanding tasks beyond machine translation, influencing subsequent developments in the field.

ChatGPT is built on the GPT-3.5 framework, a more compact iteration of the GPT-3 model released by OpenAI in 2020. Although GPT-3.5 has 6.7 billion parameters, compared to GPT-3's 175 billion, it maintains high efficacy in various NLP tasks, including language understanding, text generation, and machine translation (Alkaissi & McFarlane, 2023; Borji, 2023).

The training of ChatGPT involved a comprehensive corpus of text data, finely tuned to specialise in generating conversational responses. This specialised training enables ChatGPT to produce responses that closely mimic human conversation, making it highly effective in handling user queries with contextual relevance and linguistic coherence (Baidoo-Anu & Ansah, 2023; Cotton et al., 2023; Howard et al., 2023; Zhuo et al., 2023).

Additionally, ChatGPT integrates advanced features such as contextual understanding and nuanced response generation, enabling it to engage in conversations requiring deeper contextual and linguistic comprehension. The design of ChatGPT includes continuous learning and adaptation mechanisms, allowing it to evolve and improve as it interacts with users and new data (Ray, 2023).

In various domains, including customer service, education, and content creation, ChatGPT's potential to transform human-AI interactions is significant. Its proficiency in understanding and generating human-like text opens up new possibilities for intuitive and natural interactions between humans and machines (Dwivedi et al., 2023).

Conclusively, ChatGPT stands as a milestone in AI and NLP, illustrating the rapid advancements in these fields. Its development highlights AI's processing and generating human language capabilities and heralds new applications in diverse societal sectors.

4.2. Versions and development

The evolution of GPT models highlights the rapid advancements in AI and NLP, culminating in the creation of ChatGPT. Each version, from GPT-1 to the latest iterations, represents a leap in natural language understanding and generation capabilities.

• GPT-1 (2018): As the foundational model in the series, GPT-1 had 117 million parameters and was a significant step in language model development. It was primarily trained to predict the next word in a sequence, leveraging the Transformer architecture. GPT-1's fine-tuning ability for various tasks like language translation and sentiment analysis demonstrated the potential of language models in NLP applications (Cotton et al., 2023; Howard et al., 2023).

• GPT-2: GPT-2 marked a substantial improvement with 1.5 billion parameters. Its enhanced text generation capabilities distinguished it, producing longer, more coherent sequences. GPT-2's improved performance across various tasks showcased the model's ability to understand and generate complex language patterns, setting a new standard in language models (Qu et al., 2020; Schneider et al., 2021; Shrivastava et al., 2021).

• GPT-3: GPT-3, with an unprecedented 175 billion parameters, represented a quantum leap in language model capability. Its extensive pre-training allowed it to perform various NLP tasks without task-specific training data, distinguishing GPT-3 from its predecessors. This version demonstrated remarkable versatility and adaptability, making it a landmark in AI language models (Hewett & Leeke, 2022; Kinoshita & Shiramatsu, 2022; Lammerse et al., 2022).

• InstructGPT: InstructGPT, an iteration of GPT-3, incorporated human feedback into its fine-tuning process. This innovation allowed for more accurate and contextually appropriate responses, enhancing the model's usability in more nuanced and specific applications. It signified a shift towards more interactive and responsive AI models (Bhavya et al., 2022; Chan, 2023).

• ProtGPT2 and BioGPT: The development of specialised versions like ProtGPT2 and BioGPT illustrated the adaptability of GPT models to domain-specific applications. ProtGPT2 focussed on protein engineering, while BioGPT was tailored for biomedical text processing. These versions highlighted the potential of GPT models to contribute significantly in specialised fields beyond general language processing (Ferruz et al., 2022a, 2022b; R. Luo et al., 2022; Mardikoraem et al., 2023; B. Wang, Xie, et al., 2023).

• ChatGPT: ChatGPT, leveraging the advancements of its predecessors, was trained on a diverse corpus, including books and websites. It excelled in generating coherent and engaging conversational responses, making it a robust tool for interactive applications. The development of ChatGPT showcased the practical applicability of GPT models in everyday AI applications (Abdullah et al., 2022).

• GPT-4: The latest iteration, GPT-4, expanded the capabilities of language models to include multimodal inputs, processing both image and text. Its human-level performance on various benchmarks illustrated significant advancements in AI, setting new standards for language models in terms of versatility and complexity (Gill & Kaur, 2023; Gimpel et al., 2023; H. Luo et al., 2023). The comparison between ChatGPT 3.5 and ChatGPT 4 is shown in Table 7.

Table 7. Functionality comparison between ChatGPT 3.5 and ChatGPT 4.

Feature	ChatGPT 3.5	ChatGPT 4
Image Processing	Unable to process or understand images	Can process and understand images
Text Generation	Generates text based on text input	Generates text based on text and image inputs
Handling Prompts and Conversations	Handles shorter prompts and conversations	Handles longer prompts and conversations
Factual Error Rate	Higher factual error rate	Lower factual error rate
Response Generation Speed	Faster in generating responses	Slower due to hourly prompt restrictions
Parameter Count	175 billion parameters	170 trillion parameters
Web Browsing Capability	Not available	Enabled, allowing ChatGPT to surf the web
Error Propagation (e.g. hallucination)	Prone to errors such as hallucinations	Not specified
Adoption in Workplace for Basic Tasks	Not specified	Adopted by many workers for basic tasks
Memory Capacity	Short-term memory around 8,000 words	Short-term memory extends to around 64,000 words
Creativity and Coherence	Less creative and coherent	More creative with improved coherence
Programming Proficiency	Less proficient in complex coding tasks	More proficient in complex coding tasks
Hallucination	Not specified	Reduced hallucination (assumed)
Cost	Lower cost per 1,000 tokens	Higher cost
Multimodal Model	Yes	Yes, includes advanced parameters, enhanced memory, and steerability
Search Capabilities and Plugin Integrations	No	Yes

5. Disciplinary diversity in ChatGPT's academic applications

RQ2: In what manner has ChatGPT been employed within diverse academic and research domains?

The uniqueness of ChatGPT lies in its adeptness at language understanding and generation tasks, which was a leap from the capabilities of GPT-3. This advancement was not just a testament to the evolving architecture of the model but also a reflection of the intricate training methodologies employed. ChatGPT was trained using Reinforcement Learning from Human Feedback (RLHF), a novel approach significantly enhancing its conversational abilities (Koubaa et al., 2023).

The literature reveals an explosion of interest post the advent of ChatGPT, with over 3000 articles, reports, and news published across various platforms until March 2023. The model found its application transcending disciplinary boundaries, utilised in fields as diverse as orthopedic research to medical education, showcasing its versatile nature (Chatterjee et al., 2023).

Furthermore, ChatGPT's history is also intertwined with the continuous feedback and improvements facilitated by the community. OpenAI's proactive stance in soliciting feedback and making iterative refinements to the model has been a cornerstone in its development, ensuring its alignment with the evolving needs and ethical standards of the user base (Deeper Insights, 2023).

The trajectory of ChatGPT's development showcases the dynamic and collaborative nature of advancements in AI and NLP technologies. It represents the fusion of sturdy architectural frameworks, advanced training methodologies, and an engaged community that drives the progress of AI. The evolution from GPT-3 to ChatGPT, and the subsequent growth of research and applications, illustrate the dynamic and collaborative nature of AI development, setting a precedent for future endeavours in this field.

5.1. Disciplinary distribution

This comprehensive study systematically classified various academic disciplines into broader categories, establishing a clear mapping between specialised subfields and their overarching domains. A key focus of our research was the integration of ChatGPT, a state-of-the-art Large Language Model (LLM), across these diverse areas. By conducting extensive case studies within the ChatGPT framework, we sought to thoroughly explore and analyse this technology's multifaceted applications and implications in various academic fields (Nawaz et al., 2023).

To elucidate our methodology, we have meticulously categorised major academic fields, demonstrating the application of ChatGPT within research articles. This categorisation is visually represented in Figure 5, offering readers an intuitive understanding of the intersections between ChatGPT and various disciplines. This graphical representation is a foundational reference point, facilitating a deeper engagement with the subsequent discussions presented in this paper.

Figure 5. Major category of discipline overlap by AI-ChatGPT.

Additionally, we employed the PRISMA methodology to conduct a robust analysis of the most commonly used author keywords in the collected data. This analysis sheds light on the prevalent terminologies within the field and highlights the emerging trends and focal points in current research. The findings from the keywords analysis are detailed in Figure 6 and further explained in Table 8, this analysis has been carried out in biblometric software tool (VOSviewer, 2023).

Figure 6. Frequent authors keyword occurrence in research articles.

Table 8. Most frequent author keywords.

Clusters	keywords	Occurence
Cluster 1	Academic integrity	25
	Chatgpt	801
	Chatbots	25
	Deep learning	19
	Ethics	17
	Generative AI	46
	Learning	108
	Plagiarism	18
	Prompt engineering	12
	Student	19
	Text generation	12
	Trust	11
Cluster 2	Conversational agent	12
	Medicine	32
	Patient information	12
Cluster 3	Generative pre	26
	Sentiment analysis	13
	Transformer	52
Cluster 4	AIchatbot	11
	healthcare	20
Cluster 5	Case report	12
	Health care	10
Cluster 6	Patient education	12
	Medicine	32

Furthermore, our investigation extended to distributing research papers across different disciplines employing ChatGPT. This aspect of the study yielded significant insights into the practical application domains of ChatGPT. Notably, our findings reveal that the healthcare sector emerged as the predominant field of application, closely followed by education, research, computer science, and information technology. A comprehensive breakdown of this distribution is presented in Figure 7, offering a panoramic view of ChatGPT's integration across various academic spheres.

Figure 7. Distribution of publication across broad discipline categories.

This study aims to provide a balanced and academically rigorous overview of ChatGPT's interdisciplinary applications, serving as a valuable resource for researchers and practitioners a like in navigating this evolving landscape.

5.2. Case studies

In this section, the article delves into various case studies related to ChatGPT research in diverse research fields. After analysing the data, we identified the top 10 authors based on their number of publications. The analysis shows that researchers are at the forefront due to their many publications. Most top researchers are in the healthcare field, giving us useful insights. For more details, please refer to Figure 8. We considered the WoS and Scopus citations here.

Figure 8. Top 10 researchers across various disciplines.

5.2.1. Health care

The AI-powered chatbot, GPT, has significantly impacted the healthcare industry, particularly in healthcare-related publications where researchers have gained valuable insights from multiple case studies. In our analysis, we have identified the top 10 researchers in this field, as shown in Figure 9. We have also highlighted their studies that exhibit more diversity in their case studies.

Figure 9. Top 10 authors in a healthcare-related number of publications.

• Rhinoplasty Consultation Simulation: ChatGPT demonstrated the capability to deliver clear and comprehensive answers to simulate an initial rhinoplasty consultation. Though it stressed individualised care, it was limited in giving detailed, personalised advice. This suggests ChatGPT could be a preliminary information source for patients, especially in areas with limited medical advice availability (Xie, Seth, Hunter-Smith, Rozen, Ross, et al., 2023).

• Radiology Residents' Information-Seeking: The study among Indian radiology residents revealed that while online resources are extensively used, ChatGPT was not a preferred tool due to its lack of images and references. It shows that for image-centric disciplines like radiology, ChatGPT's current version falls short. However, its potential for non-medical professional education and report templating was recognised (Sethi et al., 2023).

• Orthopaedic Education during COVID-19: ChatGPT-4 was found to provide accurate medical advice for orthopedic scenarios, offering a valuable educational supplement during times of reduced clinical exposure. Its limitations in specialised fields were noted, but its overall proficiency indicates a promising role in medical education (Lower et al., 2023).

• Breast Augmentation Patient Education: ChatGPT-4 was evaluated for its ability to provide safe and up-to-date medical information on breast augmentation. It performed well in constructing responses but showed limitations in personalisation and sometimes provided inappropriate references. This underscores the need for further development to enhance its reliability as a patient education tool (Xie, Seth, Rozen, et al., 2023).

• Aesthetic Plastic Health Care Research: In aesthetic plastic health care research, ChatGPT provided relevant information but was criticised for creating fictitious references. This underscores the importance of oversight when using AI in academia to ensure academic integrity (Seth, Cox, et al., 2023).

• Clinical Decision-Making Support for Junior Doctors: Comparing ChatGPT with other LLMs, it provided more comprehensible and clinically aligned advice. This suggests that ChatGPT could enhance self-directed learning and decision-making support, though further development is needed for its integration into medical education (Xie, Seth, Hunter-Smith, Rozen, & Seifman, 2023).

• Augmenting Medical Research Writing: A conversation with ChatGPT about thumb arthritis indicated that while ChatGPT-3 can provide accurate basic information, it lacks depth in analytical abilities and reference provision. This suggests a cautious approach to its use in medical publishing (Seth, Kenney, et al., 2023).

• Patients' Education on Dermatology: ChatGPT could generate easily understandable text for patient education on common dermatological diseases. The readability was suitable for high school to college-level understanding. However, the similarity index was higher than desired, indicating potential issues with text originality (Mondal et al., 2023).

• Higher-Order Reasoning in Pathology: ChatGPT showed proficiency in answering higher-order reasoning pathology questions with an 80% accuracy level. The responses were consistently in the relational category according to the SOLO taxonomy, which indicates meaningful connections within the text (Sinha et al., 2023).

• Drug-Drug Interactions (DDIs): ChatGPT was partially effective in predicting and explaining DDIs, which is crucial for patient safety (Juhi et al., 2023). While most responses were correct, a significant number were inconclusive, highlighting the need for further refinement before relying on AI for DDI information.

• Case Vignettes in Hematology: ChatGPT outperformed Google Bard and Microsoft Bing in solving hematology-related cases. It scored higher than the 50% threshold, suggesting its potential usefulness in medical education (Kumari et al., 2023).

• Case Vignettes in Physiology: ChatGPT again led the performance, indicating its superior capability in processing and answering physiology case vignettes. Such tools can be considered for case-based learning in physiology, but further evaluation is needed.

• Microbiology Education: ChatGPT demonstrated an 80% accuracy rate in answering both first- and second-order knowledge questions in microbiology (Das et al., 2023). Performance varied across different microbiology topics, suggesting some inconsistencies.

  Triage in Emergency Rooms for Metastatic Prostate Cancer:

• ChatGPT was tested for its sensitivity and specificity in triage decisions. The AI demonstrated a high sensitivity to inpatient admission decisions and provided accurate diagnoses and additional treatment recommendations. The study suggests ChatGPT could enhance triage and care quality in emergency settings (Gebrael et al., 2023).

• Diabetes Patient Education: The necessity of continual patient education in diabetes care is highlighted, and ChatGPT's potential to provide tailored support is recognised. The post emphasises the importance of ethical and equitable development of AI tools like ChatGPT for diabetes education (Khan & Agarwal, 2023).

• Reasoning-Based MCQs in Medical Physiology: ChatGPT, Bard, and Bing were evaluated on their ability to generate reasoning-based multiple-choice questions for medical students. Although AI tools showed some ability to create valid questions, significant differences were noted among the models, indicating room for improvement (Dhanvijay et al., 2023).

• Uveitis Diagnosis Comparison: The diagnostic performance of ChatGPT and Glass was assessed against uveitis specialists. ChatGPT demonstrated promising results, though not as accurate as the specialists. Clinicians showed positive attitudes towards integrating AI in healthcare practice (Rojas-Carabali et al., 2023).

• Evaluation of ChatGPT in Medical Query Resolution: A cross-sectional study with physicians generated medical questions to assess ChatGPT's accuracy and completeness in responses. The AI provided accurate and complete information, with better performance over time and between updated versions. The need for further research and model improvement is emphasised for clinical application (R. S. Goodman et al., 2023b).

• ChatGPT in Managing Snakebite Enquiries: ChatGPT was evaluated for its effectiveness in providing information on venomous snakebites. The AI model gave accurate responses but showed limitations in personalisation and regional specificity. It is a valuable resource for initial information and could help in remote areas, although it should not replace professional medical advice (Altamimi et al., 2023).

• Enhancing Medical Conference Discussions with ChatGPT-4:ChatGPT-4 was used in a Pan-Arab Pediatric Palliative Critical Care Conference to enhance discussions. It helped summarise and contrast key themes, particularly around complex DNR conflict resolution. The study suggests ChatGPT-4 can aid critical thinking in medical professionals, with further validation needed (Almazyad et al., 2023).

• Response of ChatGPT to Pediatric Intensivists During Respiratory Virus Outbreak: The utility of ChatGPT was explored in generating mitigation strategies for overwhelmed hospitals during a post-pandemic respiratory virus outbreak. ChatGPT provided suggestions deemed useful by healthcare providers but required validation and supplementation by experts. The potential for AI chatbots to support healthcare systems in rapidly changing situations was acknowledged, with a call for expert validation and further research (Alhasan et al., 2023).

• ChatGPT for Resident Education in Plastic Healthcare: ChatGPT was evaluated as a tool for resident self-evaluation using the PSITE exam questions. The AI demonstrated 54.96% accuracy and was particularly adept at using logical reasoning and internal information. The conclusion supports using ChatGPT for resident education, noting its potential to enhance learning and promote evidence-based medicine (Gupta et al., 2023).

• ChatGPT in Clinical Practice: ChatGPT is used for various purposes, including generating differential diagnosis lists, optimising clinical decision support (CDS), and assisting in medical documentation. The technology has improved efficiency and accuracy in generating clinical letters, radiology reports, and discharge summaries. Future research is directed toward real-time monitoring, predictive analytics, precision medicine, telemedicine, and integration with healthcare systems. The viewpoint stresses the importance of understanding both the benefits and potential dangers of using ChatGPT in clinical practice (J. Liu, Wang, et al., 2023).

• Ethical Challenges of ChatGPT in Healthcare: Ethical issues with ChatGPT arise from legal, humanistic, algorithmic, and informational concerns, such as liability, patient privacy, physician-patient relationships, and integrity. The tool also raises concerns regarding algorithmic bias and transparency, as well as the validity and effectiveness of the information provided. Rigorous validation and updates based on clinical practice are essential to ensure accuracy and reliability. Ethical guidelines are necessary for the responsible use of AI in healthcare to maintain patient trust and enable informed decision-making (C. Wang, Liu, et al., 2023).

• Improving CDS Logic with ChatGPT: This study assessed whether ChatGPT can offer useful suggestions for improving CDS logic and how it compares to human-generated suggestions. ChatGPT-generated suggestions were unique, understandable, and relevant, with moderate usefulness and some concerns over acceptance and redundancy. AI-generated suggestions could complement CDS alert optimisation, helping to identify improvements and support their implementation. The potential for using LLM in complex clinical logic is recognised as a step towards an advanced learning health system (S. Liu, Wright, et al., 2023).

5.2.2. Computer science and IT

The field of AI is closely related to computer science. Researchers have integrated it into various subfields of computer science and IT, such as communication networking and programming. In this regard, we have highlighted the top 10 authors who have conducted research related to computer science and IT, as shown in Figure 10.

Figure 10. Top 10 researchers in CS and IT.

Integration of AI in Intelligent Vehicles: The application of ChatGPT and similar AI technologies in intelligent vehicles is a burgeoning field. The discussion may focus on how conversational AI can be incorporated into vehicle systems to enhance human-machine interaction, potentially leading to advances in autonomous driving capabilities and intelligent transportation systems (J. Zhang, Pu, et al., 2023).

• Computational Challenges: The high computational costs of running sophisticated AI models like ChatGPT in real-time environments, such as those required for intelligent vehicles, pose significant challenges. Discussions might delve into optimising computational resources and developing more efficient AI models that can operate within the constraints of vehicular technology (F.-Y. Wang, Li, et al., 2023).

• Human-Machine Augmentation: The idea of augmenting human-machine interaction with AI offers exciting possibilities for the future of intelligent vehicles. A discussion could explore how human-machine-augmented intelligent vehicles (HiVeGPT) could balance control and decision-making between AI systems and human drivers, ensuring safety and efficiency (J. Zhang, Pu, et al., 2023).

• Ethical and Practical Considerations: Implementing AI in sensitive applications such as vehicles brings ethical and practical considerations to the forefront. Discussions might touch on the responsibility for AI-driven decisions, privacy concerns related to data collection and usage, and the integrity of the AI systems in terms of providing unbiased and accurate information (F.-Y. Wang, Li, et al., 2023).

• Scenario Generation and Decision Making: The complexity of scenario generation for intelligent vehicles and the uncertainty of AI decisions are critical issues. The discussion would likely examine methods for creating realistic driving scenarios for AI testing and the reliability of AI in making safe driving decisions (Gao et al., 2023).

• Framework Proposals: Proposing new frameworks for integrating AI into intelligent vehicles is a key topic of discussion. The conversation would focus on the design, feasibility, and implementation of these frameworks, as well as their potential to facilitate the development of intelligent vehicle systems (J. Zhang, Pu, et al., 2023).

• Updates and Advancements: Keeping AI models like ChatGPT up-to-date with the latest information and integrating them with existing systems is an ongoing challenge. Discussing this topic would involve strategies for continuous learning, real-time updating, and predictive analytics within the AI models (Du et al., 2023).

• Impact on Research and Development: Finally, the impact of AI on R&D within the intelligent vehicle space is a significant point of discussion. The conversation might cover how AI technologies can accelerate the pace of innovation, the types of research currently being pursued, and the expected trajectory of future developments in intelligent vehicle technology (F.-Y. Wang, 2023).

• The advent of language models like ChatGPT has broadened the computational landscape, particularly in social systems and AI linguistic frameworks. Wang et al. delve into the utilisation of ChatGPT within computational social systems, examining its transition from simple conversational applications to more complex, human-oriented operating systems (F.-Y. Wang, Li, et al., 2023). Their work likely discusses the integration of AI in social contexts, emphasising the need for systems that not only process information but also understand and adapt to human nuances, potentially transforming user interaction with technology.

• Kumar et al. propose a structured approach to evaluating AI linguistic models through their framework, testFAILS (Kumar et al., 2023). This framework could provide a systematic method for testing and validating the linguistic capabilities of AI systems, ensuring their effectiveness and reliability in real-world applications. Such a framework is vital for the progress and trustworthiness of AI systems, addressing challenges in consistency, accuracy, and the nuanced understanding required in language processing tasks.

• A Secure and Privacy-Preserving Blockchain-Based XAI-Justice System: The discussion likely elaborates on the use of blockchain and AI to create a justice system that is both secure and privacy-preserving. It might detail technological innovations such as NLP, ChatGPT, ontological alignment, and the semantic web. The numerical aspect of the discussion might revolve around the system's performance metrics, although specific figures aren't provided in the abstract.

• Analysing Sentiments Regarding ChatGPT Using Novel BERT: The paper's (Mujahid, Rustam, et al., 2023)discussion section would probably explore the intricacies of sentiment analysis using the BERT model, including the machine learning techniques and deep learning models compared. The numerical value mentioned in the abstract is a 96.49% accuracy rate achieved by the proposed BERT model in sentiment analysis tasks.

• Arabic ChatGPT Tweets Classification: This paper's (Mujahid, Kanwal, et al., 2023)discussion will focus on the methodology and performance of the hybrid transformer-based model used for sentiment classification of Arabic tweets about ChatGPT. Numerical values provided in the abstract include a 96.02% accuracy rate for the hybrid model, 100% precision on negative tweets, and 99% recall for neutral tweets, which are significant metrics indicating the model's performance.

• Enhancing NLP with ChatGPT's Novel Knowledge: The author of Amin et al. (2023a) would detail how ChatGPT's verbose output contributes unique insights when fused with specialised models. The abstract suggests that ChatGPT possesses novel knowledge beneficial for NLP tasks, which, when fused (either early or late fusion), enhances performance on affective computing problems. Opportunities: By leveraging the novel knowledge of ChatGPT, there is potential to create more nuanced models that can perform a wider array of tasks with improved accuracy. Challenges: Identifying optimal fusion techniques and quantifying the degree of improvement pose challenges, especially in the absence of specific statistical measures in the abstract.

• Comparative Analysis of Generalist and Specialist Models: This paper (Amin et al., 2023b) comparison of ChatGPT's performance to other models would focus on its capability as a generalist tool. While the RoBERTa model performs better for specific tasks, ChatGPT achieves comparable results to Word2Vec and BoW, demonstrating its versatility. ChatGPT's robustness in noisy data scenarios is a highlight, as indicated by the lesser performance of Word2Vec in such conditions. Opportunities: ChatGPT's ability to handle noisy data opens up use cases in environments where data preprocessing is limited, offering a practical solution in less controlled settings. Challenges: The limitation here is the absence of specific statistical outcomes for ChatGPT's performance relative to the Word2Vec and BoW models in the abstract, which would have provided a clearer benchmark for comparison. Evaluation of ChatGPT's Text Classification Capabilities: The abstract points to an evaluation of ChatGPT against three baselines: RoBERTa-base, Word2Vec, and BoW, across three affective computing tasks. Decent results across these tasks confirm ChatGPT's generalist nature, although it does not surpass the RoBERTa model trained specifically for these tasks. Opportunities: ChatGPT's broad applicability could be especially beneficial in developing low-resource applications where extensive model training isn't possible. Challenges: The challenge here is the abstract's lack of numerical performance data for ChatGPT versus the baselines, which would have helped determine the practicality of using ChatGPT in various applications.

• Impact of AI-Generated Fake Information on User Perceptions: The paper (Amaro et al., 2023) would delve into the implications of fake information produced by AI on user trust and satisfaction. The user study on ChatGPT revealed that early exposure to fake information led to a significant decrease in trust and satisfaction among users compared to those who encountered such information later or not at all. A statistically significant difference was found in trust and satisfaction when users encountered fake information early on. However, no significant difference was noted between the late-exposure group and the control group.

  Opportunities: These findings suggest an opportunity to educate users about the potential for AI to generate incorrect information, promoting critical and informed usage of AI tools. Challenges: The challenge lies in managing user expectations and maintaining trust in AI systems, especially when users are confronted with fake information at the outset.

• Effect of Timing on the Perception of AI Reliability The research investigates how the timing of exposure to fake information from ChatGPT affects user trust. Early knowledge of AI fallibility has a different impact than later, with early exposure diminishing trust more significantly. The experiment, involving 62 university students, showed higher usability and net promoter score (NPS) when fake information was detected later in the interaction.

• Strategies for Enhancing Critical Engagement with AI Tools: The research work also points toward the need for strategies that encourage critical engagement with AI tools such as ChatGPT. This could involve teaching users to verify information provided by AI. Despite a decrease in trust and satisfaction after encountering fake information, the perception remained high, suggesting that some users still see value in using ChatGPT, albeit with a more cautious approach.

  In summary, while ChatGPT can be a valuable tool for various tasks, its tendency to occasionally produce fake information can impact user trust and satisfaction. This impact is modulated by when users become aware of this limitation. The key opportunity lies in using this awareness to foster a critical and informed approach to AI use, which can also serve educational purposes. However, this also challenges maintaining user trust and engagement with AI systems.

• Performance of ChatGPT in Named Entity Recognition and Classification (NERC) Tasks: The author in W. Zhang, Wang, et al. (2023) would revolve around the exploration of ChatGPT's ability in NERC tasks in primary sources, such as historical texts, in comparison with state-of-the-art language models.

  Statistical Outcomes: Although specific numerical performance metrics are not provided in the abstract, it mentions several shortcomings in entity recognition, which indicates areas where ChatGPT's performance is lacking compared to other models. Opportunities: ChatGPT's “zero-shot” capabilities for generating plausible-sounding answers present an opportunity to extend its application to historical document analysis and other specialised domains.

  Challenges: Challenges include inconsistency in entity annotation guidelines, entity complexity, code-switching, and the inaccessibility of historical archives. These factors all negatively impact ChatGPT's performance in specialised NERC tasks. The discussion highlights that while ChatGPT is recognised for its potential in higher education and specialised tasks like NERC, it faces challenges related to its reliability and consistency, especially in domains where data is scarce or complex. The numerical data provided underscores the need for further research and development to mitigate these issues and harness the full potential of ChatGPT in academic and research settings.

5.2.3. Engineering and technology

• The researchers have integrated ChatGPT across multiple engineering disciplines to perform various tasks tailored to their specific needs. These disciplines include biomedical, mechanical, and chemical engineering, among others. Through our analysis, we have identified the top 10 authors in the main engineering and technology disciplines, as illustrated in Figure 11.

• Bio-medical engineering: Cheng, Li, et al. (2023), this author used GPT for Biomedical engineering, an interdisciplinary field that applies engineering principles and design concepts to medicine and biology for healthcare purposes. This research likely explores the engineering challenges and solutions in developing, testing, and implementing AI tools in the healthcare environment.

  Enhanced Data Analysis: AI can process vast amounts of data quickly, improving the identification of disease patterns and treatment outcomes.ChatGPT could assist healthcare professionals with diagnostic decision-making by providing up-to-date medical information and suggesting potential diagnoses. AI might enable personalised care plans by considering individual patient data. The model could be an educational platform for patients and healthcare workers to learn about infectious diseases. Streamlining administrative tasks in healthcare settings, potentially reducing costs and increasing efficiency.ChatGPT could be used to monitor disease outbreaks and spread in real-time, facilitating quicker responses.

• GPT-4's natural language capabilities can be harnessed in medical imaging, medical devices, bioinformatics, biomaterials, biomechanics, gene and cell engineering, tissue engineering, and neural engineering (Cheng, Guo, et al., 2023). Advancement of Medical Imaging: Enhanced image analysis could improve diagnostics and patient outcomes.

  Development of Medical Devices: Smart medical devices with AI integration could offer new levels of patient care and monitoring.

  Progress in Bioinformatics: Handling complex biological data can lead to breakthroughs in gene sequencing and other bioinformatic applications.

  Breakthroughs in Biomaterials and Biomechanics: AI can simulate and model new materials and biomechanical systems.

  Enhancements in Gene and Cell Engineering: Optimizing gene editing and cell engineering processes through predictive modelling.

  Growth in Tissue and Neural Engineering: Guiding the creation of artificial organs and improving neural interfaces.

• mechanical engineering: The paper (Badini et al., 2023) presents a study on applying ChatGPT, a Large Language Model (LLM) by OpenAI, to optimise the Gcode generation process in Additive Manufacturing (AM), commonly known as 3D printing. The research focuses on using ChatGPT to enhance the Gcode generation, which is pivotal in controlling the 3D printer's operations and ensuring high-quality print outcomes with minimal time and material waste. The paper indicates that performance tests have been conducted, assessing ChatGPT's ability to evaluate printing parameters and troubleshoot common issues in Fused Filament Fabrication (FFF) with thermoplastic polyurethane. The results suggest that ChatGPT has the potential to significantly impact AM by streamlining the Gcode generation process and troubleshooting, thereby making it more efficient, cost-effective, and user-friendly.

• Chemical engineering: The paper (Long-hao et al., 2023) discusses the advancement of AI technology and focuses on using LLMs like ChatGPT for processing information. It addresses a gap in current technology: the challenge of effectively inputting comprehensive information into these systems for processing and exchange.

  The core contribution of the paper is a novel method that combines the capabilities of LLMs with extensions, an interdisciplinary field dealing with the theory of solving contradictory problems. This method involves constructing a ”basic-element base,” which serves as a localised repository of input information that the LLM can process.

  The paper's methodological innovation is demonstrated through the application to the issue of rice waste, a significant problem with environmental and economic impacts. By applying this method, the paper claims to successfully address rice waste in a practical project, thereby verifying the method's feasibility. Overall, the paper proposes a method that leverages the strengths of LLMs to tackle complex problems by creating structured inputs from unstructured data, with the practical application to the management of rice waste serving as proof of concept.

• Electronic Engineering: The paper (Tafferner et al., 2023) evaluates the use of ChatGPT, a generative large language model, in the context of electronics research and development, particularly focussing on applied sensors in embedded systems. It does so through a case study involving the initial development tasks of a smart home project, testing ChatGPT's ability to provide detailed technical information and perform literature surveys. The paper presents a qualitative analysis, performance evaluation, and a critical discussion of ChatGPT's capabilities. It provides the queries, responses, and code to offer practical insights for electronics professionals considering AI tools like ChatGPT in their work. Despite the limitations, the study suggests that ChatGPT can still add value to electronics R&D, especially if combined with rigorous human oversight.

• Civil engineering: This research (Dai et al., 2023) area is critical for optimising renewable energy output and maintaining grid stability. The study's novelty lies in employing a model predominantly used in natural language processing (NLP) for a distinctly different domain: energy forecasting.

  The paper's methodology involves using the self-attention mechanism inherent in Transformer models to handle complex temporal relationships in large-scale time series data associated with wind power. This approach departs from traditional forecasting models, which may not efficiently capture such intricate dynamics.

  In terms of results, the Transformer-based model showcased superior performance over conventional methods, as indicated by various performance metrics. This enhanced accuracy suggests that such models hold significant promise for advancing the precision of wind power forecasting.

• Electrical engineering: Nair et al. (2023), ChatGPT's ability to interactively assist in generating software and hardware code, logic designs, and synthesising designs for FPGA and ASIC is highlighted as a significant advancement. However, the paper identifies a crucial gap: the potential security vulnerabilities in hardware code generated through unvetted ChatGPT prompts.

  The study systematically investigates strategies for designers using ChatGPT for secure hardware code generation. This involves prompting ChatGPT to create scenarios listed in the Common Vulnerabilities Enumerations (CWEs) under the hardware design view from MITRE, specifically CWE-1194. The researchers demonstrate the generation of insecure code by ChatGPT based on diverse prompts and subsequently propose techniques for designers to guide ChatGPT in creating secure hardware code. The paper culminates in creating secure hardware code for 10 significant CWEs on the MITRE site.

• An AI-based robot equipped with bi-directional communication capabilities and integrated with advanced technologies (Venkataswamy et al., 2023). The humanoid doctor's primary function is to diagnose diseases in patients in real time. It leverages IoT devices and edge devices to gather patient details, which include not only physiological data but also text inputs.

  The humanoid doctor processes this data to provide diagnostic opinions. Cloud AI platforms train models on historical patient data, tested against new patient data collected via medical IoT and edge devices. The paper focuses on the disease identification process, segmented into three stages, asserting that the humanoid doctor could outperform human healthcare professionals in disease diagnosis. However, the paper acknowledges significant challenges in realising such a system. These include the need for a multi-featured, accurate model and concerns regarding accessibility, availability, and technology standardisation. The paper concludes by encapsulating patient input, AI, and response mechanisms to actualise the concept of a humanoid doctor.

Figure 11. Top 10 authors in engineering and technology-related number of publications.

5.2.4. Education and research

• The paper (Vrontis et al., 2023) investigates the role of ChatGPT and skilled employees in business sustainability, particularly in glomerular disease. It explores the adoption of applications supported by GPT-3.5 and GPT-4 platforms amid a shortage of trained workers. The research uses a theoretical model validated through the PLS-SEM technique based on 209 respondents' feedback, finding significant impacts of ChatGPT and skilled workers on business sustainability, with leadership motivation also playing a crucial role.

• The paper (Khademi, 2023) delves into the transformative potential of LLM AI in educational settings, with a focus on prompt engineering as a mechanism to unlock LLM AI's capabilities. The paper discusses how LLM AI can revolutionise educational practices, emphasising the importance of critical thinking skills and the role of educators and students as active participants in learning. It calls for further research on prompt engineering's efficacy and the need to address the possible defects, biases, and ethical concerns of LLM AI in education.

• The authors Mohammed et al. (2023) describe a protocol for developing, validating, and utilising a knowledge, attitude, and practice (KAP) assessment tool towards ChatGPT in pharmacy practice and education. The method includes a comprehensive literature search, content validation, face validation, readability tests, reliability assessment, and exploratory factor analysis. The study aims to conduct KAP surveys in LMICs using the validated tool to understand the impact of ChatGPT on pharmacy practice and education.

• The study (Rao et al., 2023) investigates the role of ChatGPT-3.5 and GPT-4 in radiologic decision-making, specifically for breast cancer screening and breast pain. The models' responses were evaluated against the American College of Radiology (ACR) Appropriateness Criteria using two prompt formats: open-ended (OE) and select all that apply (SATA). In breast cancer screening, both ChatGPT-3.5 and GPT-4 scored an average OE of 1.830 out of 2. ChatGPT-3.5 achieved 88.9% correctness in SATA format, while GPT-4 scored 98.4%. For breast pain, ChatGPT-3.5's OE score was 1.125, with a 58.3% SATA correctness, compared to GPT-4's OE score of 1.666 and 77.7% SATA correctness. These statistics demonstrate a clear quantitative advantage for GPT-4 over GPT -3.5 in aligning with ACR guidelines. Emphasizes the measurable progress in AI performance from GPT-3.5 to GPT-4, suggesting a trend towards more accurate AI support in medical imaging and decision-making processes. It emphasises the measurable progress in AI performance from GPT-3.5 to GPT-4, suggesting a trend towards more accurate AI support in medical imaging and decision-making processes.

• The paper (Habibi et al., 2023) explores the integration of ChatGPT and other natural language processing models into higher education, addressing practical and ethical considerations. It highlights a gap in understanding students' perspectives on the use of ChatGPT. An exploratory study using semi-structured interviews with university students provides empirical insights into their views on this technology. The analysis identified three main themes: (1) support for autonomous learning, suggesting that ChatGPT can aid students in learning independently; (2) digital and artificial tutoring, indicating a potential role for ChatGPT as a tutoring tool; and (3) academic misconduct and ethical considerations, raising concerns about the potential misuse of ChatGPT in academic settings.

• The authors Y. Zhang, Towey, et al. (2023) present a pilot study on the use of ChatGPT for automating the generation of metamorphic relations (MRs) in the context of autonomous driving systems (ADSs) testing. The study addresses the oracle problem in ADS testing–the difficulty is in verifying the correctness of a system's output–by employing metamorphic testing (MT). It explores the potential of ChatGPT to streamline the MR generation process, which is traditionally manual, time-consuming, and prone to errors. A detailed methodology was developed for generating MRs using ChatGPT, and the quality of the MRs produced was evaluated based on domain knowledge and existing MRs. The findings suggest that ChatGPT can effectively create high-quality MRs, thus considerably reducing the manual effort involved in MR generation. The results indicate an enhancement in efficiency, quality, coverage, scalability, and reusability in software testing for ADSs.

• In Sarma et al. (2023), the authors discuss the potential of ChatGPT, a sophisticated AI language model, to revolutionise head and neck oncology. It explores ChatGPT's utility in clinical settings, such as scheduling, diagnosis, treatment planning, follow-up, telemedicine, medical documentation, scientific writing, and research. It emphasises the model's capability to assist oncologists and contribute to patient care, education, and research, thereby improving clinical outcomes.

• The article reviews the advent of Generative AI in materials science, detailing its progression from task-specific models to more generalised applications thanks to developments like the Prompt paradigm and reinforcement learning from human feedback (RLHF). It examines the structure-activity relationships within the field, evaluates the methodologies of various generative models, and investigates the practical uses of GAI, including materials inverse design and data augmentation. Using ChatGPT as a case study, the paper delves into the broad potential of GAI, such as generating materials-related content, solving differential equations, and answering frequently asked questions about materials. It presents an in-depth analysis of the advantages and limitations of these models in materials science applications. The study identifies six major challenges encountered in applying GAI to materials science, such as data quality and model interpretability. It offers potential solutions to these challenges, aiming to foster effective and explainable materials data generation and analysis methodologies.

• The research in Tyson (2023) critically examines the limitations of ChatGPT, particularly in its application to chemical education. It highlights key shortcomings, such as the inability to perform mathematical operations reliably, make conceptual errors, and generate partially accurate but plausible-looking citations. The paper reflects on the role of ChatGPT as an assistant in literature search, specifically in the biogeochemistry of arsenic, and critiques its effectiveness as a chemistry tutor.

  Critical Observations and Recommendations: The author emphasises the potential risks of using ChatGPT in educational settings, especially chemistry, due to its inaccuracies and conceptual errors. There's a call for caution in the enthusiasm for ChatGPT's application in chemical education, urging educators and students to rely on more reliable resources like Google, Wikipedia, and scholarly databases.

  Survey Results and Student Use: A survey conducted at the University of Massachusetts Amherst in April 2023 received responses from 396 students across STEM majors. The results showed that 40% had used ChatGPT, with their usage categorised as personalised learning support (48%), research support (8%), creative thinking support (26%), writing (23%), reading (13%), assessments (3%), coursework (12%), personal non-coursework (12%), and playing/exploring (3%).

• The article (Ratten & Jones, 2023) discusses the transformative impact of ChatGPT on management education, focussing on the challenges it presents in assessments and grading. It highlights ChatGPT's untraceable use, creating dilemmas for educators who seek to incorporate this technology while ensuring authentic learning. The need for rapid policy implementation regarding ChatGPT and similar AI tools is emphasised due to their ease of use and affordability. The paper contributes to understanding how technological innovations like ChatGPT can be integrated into curriculum design and management learning practices, calling for a re-examination of current educational strategies to effectively incorporate these innovations. ChatGPT is recognised as a significant technological innovation that is reshaping curriculum development and learning practices in management education. This article is one of the first to examine ChatGPT's role specifically in this context. Given the large number of students in management education and its global economic importance, understanding ChatGPT's implications is crucial. The article suggests ways for management educators to integrate AI into assessments and teaching, helping those struggling to find a balance between the benefits and drawbacks of generative AI like ChatGPT. It calls for further research to develop effective assessment practices in the ChatGPT era, including the design of assessments that can counteract the challenges posed by ChatGPT. The article also notes the increased pressure on e-learning programmes to verify student identities due to the emergence of ChatGPT.

• The research explores the capability of ChatGPT, a generative pre-trained transformer model, to complete a sophomore-level digital design laboratory course (CEC222) at Embry-Riddle Aeronautical University. The study involved ChatGPT describing the wiring, writing programmes, and answering questions for a set of existing laboratory assignments. ChatGPT's performance was evaluated by independently grading its work.

  Study Findings and Performance: ChatGPT achieved a 73% grade on the laboratory assignments, a score comparable to that of a typical student. The success of generating appropriate prompts was contingent on a solid understanding of the related lecture course material. The results suggested that ChatGPT's performance was on par with many students, potentially positioning it as a peer in a peer-to-peer learning context.

• The paper (Borger et al., 2023) discusses the impact of LLMs and AI technologies like ChatGPT and Bard on scientific research and education. It highlights a symposium at the Walter and Eliza Hall Medical Research Institute (WEHI) that focussed on the practical applications of LLMs in medical research and the emerging ethical, legal, and social implications. The symposium, involving early career researchers, lab heads, educators, and policymakers, explored diverse topics like AI's role as an editor for scientific papers and grants, its ability to empower non-coders in bioinformatics, and its proficiency in analysing big data rapidly. It also touched upon AI-driven tools like Alphafold and protein hallucination. A significant part of the discussion revolved around the broader societal implications of using AI in science. Concerns about ethics, privacy, confidentiality, and security of research data were central themes, given the sensitive nature of medical research. 

• We have concluded that ChatGPT is used more effectively for learning and education across multiple fields. Our analysis has identified several positive aspects of ChatGPT that enhance the robustness of the learning and education process. These aspects cover various disciplines for learning and education, as depicted in Figures 12 and 13, where we have also analyzed data among the top 10 authors based on their publications.

Figure 12. ChatGPT usage as a learning and education tool across all disciplines.

Figure 13. Top 10 authors in education and research-related number of publications.

5.2.5. Pyschology and social sciences

• The paper (Carvalho & Ivanov, 2023) likely discusses the transformative effects of AI, specifically ChatGPT and LLMs, on the tourism sector. It might explore how these technologies are currently applied across various stakeholder operations within the industry. The discussion could entail an analytical overview of how AI is integrated into customer service and operational efficiency, and the potential impact on employment and human resources in tourism.

• The discussion in the paper (Ivanov & Soliman, 2023) likely revolves around the transformative role of ChatGPT in tourism education and research. It probably explores how the ability of ChatGPT to generate text for assignments and research papers could change the current academic landscape. This would involve an examination of the current methods in tourism education and research and how they may be altered by the integration of AI.

• The paper (Dogru et al., 2023) likely examines the multifaceted impact of Generative AI on the hospitality and tourism (HT) industry. It offers a critical examination of GAI's effects from the perspectives of various stakeholders within the industry, including operations, design, marketing, destination management, human resources, revenue management, accounting and finance, and strategic management. The discussion would integrate both practical and academic viewpoints, aiming to provide insights and foresight into the applications of GAI.

• The author Iskender (2023) presents a unique approach by interviewing ChatGPT, OpenAI's GPT-3 model, to explore its impact on higher education, academic publishing, and specifically the hospitality and tourism industry. The discussion focuses on the role of AI-based machine learning models in these sectors and gathers insights directly from ChatGPT regarding its applications and limitations.

• The study (D'Souza et al., 2023) investigates ChatGPT 3.5's proficiency in psychiatry by evaluating its responses to 100 clinical case vignettes. The responses were appraised by expert faculty from the Department of Psychiatry, focussing on various aspects of psychiatric care, including management strategies, diagnosis, and treatment. The discussion likely revolves around the implications of ChatGPT's performance for mental health care, particularly in settings with suboptimal diagnosis and treatment.

• The research (Hung & Chen, 2023) delves into the divergent opinions on the use of ChatGPT by Chinese students in academia. It identifies the contrasting views between conservative stakeholders concerned about academic dishonesty and progressive educators advocating for the integration of AI technologies to enhance academic quality. The study also spotlights the issue of plagiarism as a significant concern for Chinese educators. The content analysis of 40 newspaper articles revealed a neutral portrayal of ChatGPT's use in Chinese academia, with a balance of positive and negative sentiments expressed in the media. The articles, particularly from March 2023, displayed a mixture of positive and negative language regarding the adoption of ChatGPT in academic learning.

• The paper (Rahimzadeh et al., 2023) debates the necessity of traditional ethics education within healthcare professional training in the presence of advanced tools like ChatGPT and other LLMs. It examines the capabilities of LLMs to aid in developing ethical competencies among future clinicians, juxtaposed against the goals of standard bioethics programmes. Through case analyses, the paper identifies how LLMs like ChatGPT align with or deviate from achieving the objectives of bioethics education. Inadequacy as a Standalone Tool: Despite their strengths, LLMs cannot fully replace human-led teaching in ethics due to the nuanced nature of ethical decision-making and the complexity of human values involved, we conducted an analysis of the top 10 authors based on the number of publications in Psychology and Social Sciences from our selected dataset, as depicted in Figure 14.

  Figure 14. Top 10 author in Psychology and social sciences.

  

5.2.6. Science

• The paper (Lee, 2023) presents a mathematical exploration of hallucination and creativity in generative pre-trained transformer (GPT) models. It involves a detailed analysis using probability theory and information theory to define and quantify these phenomena. The focus is on understanding the interplay between hallucination (where the model generates inaccurate or irrelevant information) and creativity (the ability to produce novel and contextually appropriate content) in GPT models. The paper establishes clear mathematical definitions for both hallucination and creativity within the context of GPT models. A parametric family of GPT models is used to explore the relationship between hallucination and creativity, identifying a balance that optimises model performance. The analysis identifies an optimal balance point where the model's performance is maximised, considering both its creative capacities and tendency to hallucinate.

• The paper (Lubiana et al., 2023) discusses the effective utilisation of ChatGPT and other LLM chatbots in the field of computational biology. It emphasises the potential of these tools to enhance productivity and streamline complex workflows, particularly for tasks that are repetitive or minor. The discussion likely revolves around best practices for integrating these technologies into bioinformatics work, balancing their benefits with an understanding of their general-purpose nature and limitations.ChatGPT can significantly improve efficiency in computational biology, especially in automating repetitive tasks. While ChatGPT is a valuable tool, overreliance can be detrimental, and it should not replace fundamental bioinformatics skills or critical thinking.

• This study (Li et al., 2023) investigates the effectiveness of ChatGPT as a tool for assisting middle school students in mathematics, particularly in flipped classroom settings. It focuses on ChatGPT's potential to enhance self-regulated learning, a critical aspect of students' academic development, especially when engaging with online resources at home. The study's core involved testing ChatGPT's accuracy in solving questions from Taiwan's past education examinations, spanning various major areas of mathematics education. High Accuracy Rate: ChatGPT achieved a high accuracy rate of 90% (graded A+) in answering math examination questions. Broad Coverage: Unlike other studies focussing on single units or courses, this research found that ChatGPT's accuracy exceeded 80% (graded A) across all six major areas of mathematics education in Taiwan. Potential as a Learning Aid: The results suggest that ChatGPT can significantly improve self-regulation issues in students, making it a valuable tool for middle school math education.

5.2.7. Business and economics

In this section and upcoming sections, all authors have made a single article contribution in this field from our extracted data. That's why We can't rank the top authors in terms of publications in this field.

• The article (Frederico, 2023), discusses the initial applications and impacts of ChatGPT in the field of supply chain management. It addresses the gap in current literature regarding ChatGPT's specific use in this domain and provides a foundational understanding for practitioners and researchers. The study is based on an analysis of content from specialised magazines, blogs, and company websites, as a systematic literature review was not feasible due to the scarcity of academic papers on the topic.

  Versatile Applications in Supply Chain: ChatGPT shows potential for diverse applications within supply chain management, including route optimisation, predictive maintenance, order shipment, enhancing customer and supplier relationships, data analysis, streamlining the ordering process, automating invoices, reducing waste, and aiding workforce training and guidance.

  Benefits for Supply Chain Management: The initial evidence suggests that ChatGPT can lead to cost reductions and improvements in overall supply chain performance, although it may take time for the technology to reach full maturity in this context.

• The paper (Cui, 2023) investigates the potential impact of ChatGPT on the growth strategies of European small and medium-sized enterprises (SMEs) in the service sector. It focuses on how AI, particularly ChatGPT and its latest iteration, GPT-4, can address various challenges faced by these enterprises. The study is grounded in qualitative research, utilising semi-structured interviews to gather insights into the ways ChatGPT can optimise and assist SMEs across multiple operational categories.

  ChatGPT has diverse applications in SMEs, impacting areas such as human resource management, strategic decision-making, fundraising, service R&D, finance, marketing, sales, administration, and operations. Significant Impact on SME Growth: The analysis of 293 items collected from interviews suggests that ChatGPT offers substantial optimisation and assistance for SMEs, contributing to their growth and development in the service sector.

  Strategic Reference for SMEs: The findings provide a comprehensive and systematic set of solutions that can serve as a strong reference for SMEs looking to leverage ChatGPT technology for optimisation and growth.

• The article examines the transformative role of Generative AI, particularly ChatGPT by OpenAI, in the business and finance sectors (Chen et al., 2023). It provides an overview of the recent advancements in generative AI, its practical applications, and the emergence of new tools in these fields. A significant focus of the study is on the use of ChatGPT to analyse corporate sentiments towards environmental policy from financial statements, assessing its impact on decision-making in financial markets. The study demonstrates that ChatGPT can effectively capture corporate sentiments from financial statements. The sentiment scores generated are shown to predict firms' risk management capabilities and stock return performance. The paper summarises various applications of generative AI in business and finance, offering examples of the latest tools and technologies.

• The article explores the integration of AI-powered chatbots like ChatGPT into entrepreneurship education, focussing on their potential uses and implications in higher education (Vecchiarini & Somià, 2023). It examines how ChatGPT can transform traditional teaching methods and activities within entrepreneurship courses, such as idea generation, business model creation, business plan writing, and conducting customer interviews. The study also assesses students' awareness, usage, and perceptions of ChatGPT, alongside identifying effective integration strategies.ChatGPT is seen as a tool that can increase efficiency in various aspects of entrepreneurship education. Supporting Creativity: The AI's capabilities can foster certain types of creativity among students, particularly in idea development and business planning. Student Perspectives: The survey reveals insights into how students perceive the benefits and limitations of ChatGPT in their entrepreneurship courses.

• The study delves into the darker side of ChatGPT's utilisation, specifically how its advanced natural language processing capabilities can be exploited by threat actors in cybersecurity (Chowdhury et al., 2023). It focuses on the use of ChatGPT in generating attack vectors, particularly in the context of phishing attacks, where it can produce realistic communications that deceive users into downloading malware or disclosing sensitive information. The study assesses the effectiveness of ChatGPT's security measures and how threat actors can potentially circumvent these controls. It also examines the implications of such abusive use on the supply chain management of attack vectors. ChatGPT's ability to generate human-like responses is being exploited to create convincing phishing communications. Supply Chain Management of Attack Vectors: The study shows how ChatGPT enhances the efficiency and effectiveness of attack vectors' supply chain management. Security Measures and their Limitations: While ChatGPT has inbuilt security controls to prevent malicious use, threat actors have found ways to bypass these measures.

5.2.8. Arts and humanities

• The authors discuss the utilisation of AI, specifically ChatGPT, in the context of teaching self-defense within the arts and humanities domain (Harasymowicz, 2022). The core of the discussion is a structured dialogue between ChatGPT and an expert in the martial arts sub-discipline. This dialogue, comprising 15 questions formulated by the expert, delves into the theoretical, legal, and ethical foundations of teaching self-defense. Through ChatGPT's responses, the paper examines the AI's state of knowledge in these areas, highlighting both the potential advantages and inherent limitations of AI in this field. The discussion critically evaluates ChatGPT's utility as a research tool and its applicability in enriching the work of sports and martial arts educators. It offers insights into the role of AI in scientific research and education, particularly focussing on how it can contribute to and challenge the traditional methodologies and ethical considerations in martial arts education.

• The paper (Agapiou & Lysandrou, 2023) explores a novel method for literature review in Earth observation and remote sensing in archaeology by leveraging AI and language models, specifically ChatGPT.ChatGPT was utilised to extract information on thematic topics related to Earth observation and remote sensing in archaeology. The study compares the efficiency and depth of information provided by ChatGPT against traditional bibliographic analysis methods. The research aims to understand the potential uses and limitations of AI, particularly ChatGPT, in conducting literature reviews in remote sensing archaeology.

• This paper (Leme Lopes, 2023) delves into the historiographical capabilities of ChatGPT, a cutting-edge AI chatbot developed by OpenAI. The author begins with a succinct history of AI, tracing the significant advancements leading to the creation of sophisticated tools like ChatGPT. The paper then shifts focus to a detailed analysis of interactions with ChatGPT, particularly emphasising a comprehensive “interview” conducted between February and May 2023 on historical science's theory and methodology.

  The primary objective of this “interview” was to explore ChatGPT's proficiency in understanding and discussing historiography, a key aspect of historical science. The author's preliminary findings from these interactions suggest that while ChatGPT demonstrates considerable promise in assisting historians with their research, the idea of completely AI-driven history production is still far off. The paper infers that although ChatGPT can be a valuable tool in the field of historiography, its current capabilities do not yet support the autonomous creation of historical narratives or analyses. This study contributes to the ongoing discourse on the role of AI in academic research, particularly in the domain of history and historiography.

5.2.9. LAW and policy

• The paper (Feldstein, 2023), discusses the significant, yet largely speculative, impact of generative AI on politics and governance. It emphasises the transformative potential of AI in these spheres.

  Threats to Democracies: A major challenge identified is the risk posed to democratic systems by powerful AI models controlled by private entities. These models can shape public discourse and influence democratic processes.

  Authoritarian Surveillance and Propaganda: The paper highlights how authoritarian regimes could leverage generative AI for enhanced surveillance and propaganda dissemination, potentially undermining human rights and freedom.

  Cyber Threats: The increased capabilities for criminal and terrorist groups to conduct cyber-attacks and disruptions using AI technologies are discussed as a significant concern.

  Transformation in Warfare: Generative AI is seen as a factor in transforming military strategies and operations, especially concerning the dehumanisation of lethal force. Rapid Adoption and Need for Regulation: The paper suggests that, unlike other innovations, generative AI is likely to be adopted quickly, necessitating the development of pragmatic approaches to manage associated risks.

• Intellectual Property (IP) Law and AI (Buckingham & Williams, 2023): A central theme of the paper is the readiness of current IP laws to handle the advancements in AI. It questions whether AI can legally invent or create, and if such creations can be protected under existing copyright laws.

  Case Studies and Legal Challenges: The paper discusses recent legal cases, notably those involving Dr. Stephen Thaler, to highlight the complexities and problems at the intersection of AI, inventions, and IP law. These cases question whether AI can possess the legal personality necessary to be recognised as an inventor or creator.

  Protection of AI-Generated Creations: There is a focus on the dilemma of whether AI-generated inventions and creations should receive any form of IP protection. Evaluation of Current IP Systems: The paper evaluates whether the existing IP framework is sufficient or deficient in addressing the impacts of AI, particularly in terms of incentives and regulations for both the inputs and outputs of AI systems.

• The study highlights ChatGPT's ability to generate complex texts, nearly indistinguishable from human writing, and its growing interest across various fields (King, 2023). The primary focus of the study is to assess ChatGPT's effectiveness in generating accurate, clear, concise, and unbiased information in support of scientific research. The paper compares the responses of ChatGPT-3.5 and ChatGPT-4 regarding digital school leadership and teachers' technology integration, covering specific aspects like digital leadership definition, skills, factors affecting technology integration, and its impact. Both versions of ChatGPT showed capability in providing information aligned with existing literature, with ChatGPT-4 outperforming ChatGPT-3.5 in terms of comprehensiveness.

• The paper (Tan et al., 2023) explores ChatGPT's potential in carrying out tasks traditionally performed by lawyers, particularly in providing legal information to laypeople. A framework is proposed for assessing the process of providing legal information, which includes evaluating the accuracy of the information and the ability to understand and reason about users' needs. The study involves an empirical investigation of ChatGPT's capabilities using several simulated legal cases. The results indicate that while ChatGPT may not always provide accurate or reliable legal information, it offers an intuitive interaction platform for laypeople. The research suggests that combining the capabilities of ChatGPT and systems like JusticeBot could lead to the development of flexible and accurate legal information tools.

5.2.10. Information science and library science

• The authors' Lappalainen and Narayanan (2023), details the creation of a custom chatbot named “Aisha” for Zayed University Library in the UAE using Python and the ChatGPT API. Aisha is intended to offer reference and support services to students and faculty, especially outside of regular library hours. The article examines the broader benefits and applications of chatbots in academic libraries, including a review of the early literature on ChatGPT's role in this area. This project is presented as a pioneering effort in utilising ChatGPT-based technology in the context of academic libraries, offering insights and implications for the future of AI in this field.

• The article (Lund & Naheem, 2023), addresses the growing influence of AI tools, like ChatGPT, on authorship and academic integrity in scholarly publishing. It investigates the authorship policies regarding AI of 300 top academic journals during late spring 2023. It was found that over half of these journals have specific policies and guidelines for acknowledging AI assistance in manuscript preparation. The common practice is acknowledging AI contributions in the methods or acknowledgment sections, with some journals creating special sections for AI usage. The study observes differences in AI authorship policies depending on the journal's publisher and its discipline. The findings are particularly relevant for publishers, editors, and researchers interested in understanding the evolving landscape of AI in academic publishing.

• The authors of Mannuru et al. (2023), explore the influence of Generative AI on various sectors in developing countries, assessing both positive and negative implications. Generative AI is defined as AI systems capable of generating content like text, audio, or video, distinct in its ability to produce novel outputs, unlike conversational AI which mainly provides replies. The study notes the rise in popularity of tools like ChatGPT during the Fourth Industrial Revolution and their effect on content creation. A significant focus is on the uneven accessibility of generative AI technologies in developing countries, hindered by limited technological access and infrastructure. It examines the possible effects of generative AI on economic growth, technology access, and shifts in education, healthcare, and environmental practices. It underscores the need for support and infrastructure to ensure generative AI aids inclusive development and does not exacerbate inequalities. The importance of integrating generative AI into the ongoing technological revolution in developing countries is highlighted as a key factor for progress and equitable growth.

5.2.11. communication and media studies

• The paper (Jungherr & Schroeder, 2023) discusses the increasing reliance of major media platforms like Facebook, TikTok, Twitter, and YouTube on AI to shape information environments, generate content, and interact with users.AI applications are shown to affect the public arena's key functions, such as making society visible to itself and providing spaces for public and counterpublics formation. The authors propose a framework to conceptualise and empirically examine AI's structural impact on the public arena. The paper suggests that the growing use of AI will enhance intermediary structures, granting them greater control over the public arena. The emphasis is on how the data-driven aspect of most AI applications might obscure challenges to the political status quo and hinder the assessment of AI-enabled interventions.

• The paper (Schäfer, 2023), highlights ChatGPT as a leading example of generative AI, focussing on its impact on the field of science communication. The essay suggests that the field has largely overlooked the role of generative AI, and calls for more research on public communication about AI and communication with AI, considering its “increased agency”. The need to analyse the overall impact of AI on science communication and the broader ecosystem is emphasised.

5.3. Geographical distribution

RQ 3: What is the global distribution of scholarly activity concerning ChatGPT, and which regions are at the forefront?

ChatGPT has emerged as a popular topic of research across multiple disciplines in recent years. Our collective research articles have revealed that ChatGPT has garnered significant global attention, with the top 10 countries that are highly influential in ChatGPT research being the United States, China, India, Germany, Australia, Canada, the United Kingdom, Japan, France, and Saudi Arabia. The United States tops the list with the highest number of ChatGPT research publications, followed by China and India, The detailed information is depicted in Figure 15. Additionally, our research reveals that a total of 80 countries have participated in ChatGPT research, as indicated in Figure 16. Furthermore, our analysis shows that the university with the highest frequency of occurrence in the data is King Saud University, followed by All India Institute of Medical Sciences and Taipei Medical University, among others. This information is illustrated in Figure 17.

Figure 15. Top 10 countries in ChatGPT research.

Figure 16. Participated countries in ChatGPT research across the globe.

Figure 17. Top 10 universities in ChatGPT research.

The geographic data of authors for ChatGPT research, spanning from 2022 to 2023, was sourced from the Web of Science (WoS), highlighting that LLMs are a topic of significant interest for both researchers and industries. It is a revolutionary technology that has the potential to transform human-machine interaction. AI is being integrated into different fields worldwide, and the interaction of various disciplines with AI has been enhanced with the help of ChatGPT.

Furthermore, ChatGPT has made AI more applicable to humanity, with ethical concerns being taken into account. The development of ChatGPT has paved the way for more advanced AI systems that can interact with humans naturally and intuitively. The potential applications of ChatGPT are vast, ranging from customer service to healthcare and education. Therefore, it is no surprise that ChatGPT has attracted significant interest from researchers, industries, and governments worldwide.

6. Challenges and opportunities

6.1. Challanges

We have outlined the generative AI base LLM-ChatGPT challenges across various disciplines in this study.

6.1.1. Healthcare

• Diagnostic Accuracy and Reliability: Ensuring the accuracy of generative AI in medical contexts.

• Data Privacy and Security: Safeguarding patient data in compliance with healthcare privacy regulations.

• Integration with Existing Healthcare Systems: Integrating generative AI tools without disrupting current workflows.

• Ethical Considerations: Addressing concerns like the potential dehumanisation of patient care.

6.1.2. Engineering & technology

• Technical Limitations and Innovation Balance: Addressing the limitations of generative AI in engineering applications.

• Safety and Reliability in Engineering Solutions: Ensuring generative AI solutions are safe and comply with industry standards.

• Interdisciplinary Integration: Integrating generative AI into various engineering disciplines effectively.

• Adaptation to Rapid Technological Advancements: Updating engineering practices to keep pace with generative AI advancements.

6.1.3. Arts & humanities

• Preserving Creative Integrity: Balancing the use of generative AI while preserving originality in arts and humanities.

• Interpreting Contextual and Cultural Nuances: Ensuring generative AI understands various cultural and historical contexts.

• Ethical Implications of generative AI-Generated Art: Addressing concerns about authorship and authenticity.

• Impact on Traditional Artistic and Scholarly Practices: Evaluating generative AI's impact on traditional roles in arts and humanities.

6.1.4. generative AI and public discourse

• Control Over Public Discourse: Centralized control over public discourse due to strengthened intermediary generative AI structures.

• Obfuscation of Political Challenges: The possibility of important political and social issues being overlooked due to generative AI's data-driven nature.

• Difficulty in Assessing Generative AI Interventions: Challenges in evaluating the impact of generative AI within media structures.

• Ethical and Societal Implications: Concerns about transparency, privacy, and potential manipulation due to widespread generative AI use in public arenas.

6.1.5. Accessibility and digital divide

• Accessibility and Infrastructure Challenges: Limited access to advanced technologies and inadequate infrastructure in developing countries.

• Deepening Inequalities: The risk of generative generative AI widening the digital divide and exacerbating existing inequalities.

• Need for Supportive Frameworks: The importance of developing supportive frameworks in developing countries for effective generative AI integration.

• Balancing Technological Advancement with Human Needs: Ensuring that generative AI development aligns with the specific needs and contexts of developing countries.

6.1.6. Academic and research integrity

• Diverse Policies Across Disciplines: Variation in generative AI authorship policies creating confusion and inconsistency.

• Balancing generative AI Contributions and Human Authorship: Defining the extent and nature of generative AI's contribution versus human authors.

• Ethical and Integrity Considerations: Aligning the use of generative AI tools with academic integrity and ethical standards.

• Adapting to Rapidly Evolving Generative AI Technologies: Keeping policies updated and relevant in the face of rapidly advancing generative AI technologies.

6.2. Oppurtunities

• generative AI in Communication and Information Dissemination

  	–	Enhancing Accessibility and Reach: The integration of generative generative AI like ChatGPT can significantly increase the accessibility and reach of science communication. generative AI bridges the gap between complex scientific information and public understanding by translating complex scientific concepts into more understandable language and offering multimodal communication options. This enhances the public's engagement with scientific discoveries and discussions.
  	–	Facilitating Dialogue at Scale: generative AI's ability to provide dialogical communication enables engagement with a broader audience, ensuring a wider dissemination of scientific knowledge. generative AI can foster interactive and responsive communication, catering to diverse queries and perspectives, which is critical in creating informed public discourse.

• Economic and Social Development

  	–	Promoting Economic Growth and Innovation: In developing countries, generative generative AI like ChatGPT can drive significant economic growth and innovation. By offering novel solutions and improved access to information, these generative AI tools have the potential to transform key industries, contributing to overall economic development.
  	–	Transforming Sectors like Education, Healthcare, and Environmental Management: The application of generative AI technologies can revolutionise various sectors. In education, generative AI can provide personalised learning experiences and resource management. In healthcare, generative AI can assist in diagnostics and patient care, while in environmental management, generative AI can help analyse large datasets to inform sustgenerative AInable practices and conservation efforts.

• Advancements in Academic and Research Practices Revolutionizing Research Methodologies:

  	–	ChatGPT and similar generative AI models are set to revolutionise research methodologies across various disciplines. From aiding in literature reviews to analysing complex data sets, generative can significantly expedite the research process, offering quick access to a vast array of information and facilitating new methods of knowledge creation and dissemination.
  	–	Standardizing generative's Role in Academic Research and Publication: The study and integration of generative in academic research necessitate the establishment of clear policies on AI authorship and contributions. The adoption of uniform guidelines by publishers and academic institutions will ensure consistency in generative usage across various journals and research papers. This clarity in AI's role will assist authors and researchers in effectively incorporating generative tools into their scholarly work. These sections elaborate on how AI, particularly ChatGPT, is influencing communication, societal development, and academic research, highlighting both the opportunities and transformative potential of these technologies.

• generative in Public Services and Governance

  	–	Enhancing Library Services and Public Legal Assistance: generative, particularly through tools like ChatGPT, is playing a crucial role in augmenting library services by providing efficient and timely assistance, even outside regular operational hours. This enhances the accessibility of library resources to students and faculty alike. In the realm of legal assistance, generative's user-friendly interface can simplify legal information, making it more accessible to the general public and facilitating a broader understanding of complex legal concepts.
  	–	Offering Innovative Tools for Governance, Policy-Making, and Security Strategies: generative technologies offer new dimensions in governance and policy-making. They can process vast amounts of data to inform policy decisions, potentially leading to more efficient governance. In terms of security, AI could contribute to sophisticated strategies in contexts like war and peacekeeping, leading to enhanced global security measures.

• Technological Integration in Specific Fields Applications in Historical Research, Legal Systems, and Healthcare:

  –

  In historical research, generative like ChatGPT can provide quick access to historical data and interpretations, enhancing research methodologies and offering diverse viewpoints. In legal systems, integrating generative with expert systems can enhance the accuracy and reliability of legal information tools. In healthcare, generative models are increasingly being used to improve diagnostic capabilities and patient care.

• Contributions to Fields like Textile Science, Computational Biology, and Electronics:

  –

  generative's integration extends to fields like textile science, where it can assist in data processing and visualisation, and computational biology, where it complements existing workflows by providing additional support and capabilities. In electronics, generative AI can advise on technical aspects and assist in hardware and software design flows.

• Ethical, Legal, and Regulatory Considerations Addressing the Legal and Ethical Implications of generative AI in Various Domains:

  –

  As generative AI becomes more pervasive, it's essential to address its ethical implications and ensure responsible usage. This includes considerations around privacy, data security, and the potential for bias in AI algorithms. By proactively addressing these concerns, we can safeguard against misuse and ensure AI's benefits are maximised for society.

• Adapting Intellectual Property Laws for AI-Generated Content:

  –

  The evolution of generative AI technologies like ChatGPT presents an opportunity to revisit and modernise intellectual property laws. As AI starts playing a more prominent role in creative and academic outputs, there is a need to establish clear guidelines on the ownership and rights of AI-generated content. This involves determining how traditional intellectual property laws can be adapted to acknowledge the contributions of AI, ensuring that both generative AI developers and users have clarity on the legal status of such content. Properly adjusted IP laws could foster innovation and creativity in the age of generative AI, providing a balanced framework that respects the rights of human creators while recognising the unique nature of AI-generated works.

• AI in Business and Operational Efficiency Streamlining Business Processes and Enhancing Decision-Making:

  • AI models like ChatGPT can revolutionise business operations by automating routine tasks, analysing vast data sets, and providing insights for strategic decision-making. In Small and Medium Enterprises (SMEs), generative AI tools can aid in formulating informed strategies, thereby enhancing efficiency and competitive edge. This leads to more streamlined processes, reduced operational costs, and improved productivity.

• Optimizing Operations in Sectors like Supply Chain, Tourism, and Finance:

  –

  In the supply chain sector, generative AI can automate and optimise various operations, leading to increased efficiency and reduced overhead costs. In the tourism industry, AI can offer personalised services, assist in content creation, and enhance destination management through predictive analytics. In finance, AI's capabilities can be leveraged for better pricing strategies, revenue management, and sophisticated financial analysis, contributing significantly to business growth and financial stability.

• Educational Advancements Utilizing Generative AI for Creative Educational Content and Innovative Teaching Methods:

  –

  AI, particularly tools like ChatGPT, can assist in the generation of new forms of study materials, assignments, and research papers, thereby enriching the variety and depth of educational content available to students. This not only aids in diversifying teaching methods but also introduces innovative approaches to curriculum design and student engagement.

• Augmenting Research Capacity and Assessment Strategies in Educational Institutions:

  –

  AI technologies can substantially enhance the research capacity within educational settings. They can assist in conducting literature reviews, data analysis, and even in drafting and editing academic papers. Moreover, the integration of generative AI in assessment strategies can lead to more innovative and adaptive learning environments, providing educators with tools to evaluate student performance more effectively and tailor their teaching methods to individual learning needs.

7. Conclusion

In this study, we explored the integration of AI and ChatGPT across various subdisciplines, mapping key disciplinary categories and identifying participation from 80 unique countries based on our data extraction. We examined case studies across multiple fields where generative AI has been applied, highlighting the diverse ways AI is being integrated into different domains. Additionally, we pinpointed leading authors with significant publications within and across these disciplines, as well as the foremost countries and institutions worldwide. This research underscores the challenges and opportunities posed by Large Language Models (LLMs) like ChatGPT, advancing the conversation and identifying potential paths for future interdisciplinary research. It stresses the importance of ongoing collaboration to fully leverage ChatGPT's capabilities for the advancement of science and society.

However, this study is not without its limitations. We did not categorise the articles by their impact factors or rankings in our extracted dataset. Moreover, our review was confined to major databases such as Scopus and Web of Science for the period 2022–2024, excluding other publishers. Future research will aim to broaden these searches across more databases to uncover additional data on GPT applications, thereby enriching our understanding of its extensive utility.

Acknowledgments

The authors would like to acknowledge the support of Prince Sultan University for funding the Article Processing Charges (APC) of this publication. In addition, the work of Muhammad Khurram Khan is supported by King Saud University, Riyadh, Saudi Arabia under project number (RSP2024R12).

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

Open Access Fee is funded by Prince Sultan University, Riyadh, Saudi Arabia.