An integrative decision-making framework to guide policies on regulating ChatGPT usage

Research novelty and contributions

In this study, we explore the application of the risk, reward, and resilience (RRR) framework within the specific context of ChatGPT and its implications in the higher education domain. By dissecting the intricate dynamics of risk, reward, and resilience and the underlying drivers, our research has illuminated the complexities inherent in decision-making concerning the integration of ChatGPT within educational settings. The RRR framework thus functions as a versatile and comprehensive mental model, equipping policymakers and higher education institutions to navigate the multifaceted challenges and opportunities linked to ChatGPT. It highlights the necessity of approaching risk, reward, and resilience holistically, recognizing their interdependence and potential cascading effects. Throughout our study, we have examined a spectrum of policy responses and interventions. These include legislative measures, restrictions, capacity-building initiatives, and strategies to address issues of access and equity. These diverse interventions show the multidimensional nature of ChatGPT-related challenges, requiring a careful evaluation of second-order effect and the need for balanced strategies tailored to the specific context and objectives of higher education institutions. In essence, our study furnishes a framework, roadmap, and decision-making rules for policymakers and higher education institutions to navigate the intricate terrain of ChatGPT ethics. Similarly, the unique contribution of this research lies in its adaptability, which is readily transferable, enabling any higher education institution to shape and implement academic integrity policies responsibly. Leveraging the RRR framework, examining risk, reward, and resilience, and conducting a thorough assessment of potential policy impacts, stakeholders can collectively work toward establishing a robust, inclusive, and resilient higher education system in the era of ChatGPT.

Study aim and organization

The aim of this study is to contribute toward providing guidance for discussion and decision-making concerning the utilization of ChatGPT in the education sectors, and by extension the society. Specifically, the study has three key objectives, (1) propose a decision-making process and guiding rules for RRR application, (2) identify key elements for RRR from existing studies through systematic literature review, and (3) illustrate the practical application of RRR on ChatGPT ethics conundrum. Hence, the study is structured as follows; “Study Background and Related Work” discusses the related works as well as covers a brief overview of the integrative policy framework underpinning this study; “Methodology” explains the review methodology through the systematic approach applied in the study; “RRR Application” discusses the process and guidelines of RRR application, the extracted data from the literature for risk, reward, and resilience in accordance to the integrative policy framework, the practical application of RRR in the context of ChatGPT ethics conundrum. “Discussion and Implications” discusses the research implications, and finally “Conclusion” covers the conclusion.

Review of the literature

The diverse range of studies presented in Table 1 reflects the growing interest from researchers regarding AI language models, specifically ChatGPT, across multiple domains (Dwivedi et al., 2023), such as healthcare (Jungwirth & Haluza, 2023; Masters, 2023; Rao, 2023), tourism (Carvalho & Ivanov, 2023; Ivanov & Soliman, 2023), journalism and media content (Pavlik, 2023), political orientation (Rozado, 2023), chemistry community, physics and science education (Emenike & Emenike, 2023; Cooper, 2023; Yeadon et al., 2023), architecture (Kwon, 2023), and entrepreneurship (Short & Short, 2023). Firstly, one of the themes found in the existing studies is the assessment of ChatGPT’s accuracy and performance (Geerling et al., 2023; Gilson et al., 2023; Ariyaratne et al., 2023; Dwivedi et al., 2023), from generating articles to answering questions, emphasizing the need for reliable and interpretable Generative artificial intelligence (Gen-AI) responses. Ethical and legal challenges related to AI-generated content are also a recurrent topic (Lee, 2023b; Salvagno, Taccone & Gerli, 2023; Ray, 2023; Karaali, 2023; Dwivedi et al., 2023), especially in the context of academic research and education, where concerns about authenticity and academic integrity arise. Moreover, many studies explore ChatGPT’s potential benefits and capabilities (Victor et al., 2023; Farrokhnia et al., 2023; Halaweh, 2023; Kooli, 2023; Cox & Tzoc, 2023; Carvalho & Ivanov, 2023; Jungwirth & Haluza, 2023; Cascella et al., 2023; Kolides et al., 2023; Dwivedi et al., 2023), and potential to address societal megatrends (Haluza & Jungwirth, 2023), with a focus on its role in writing, research, and pedagogy, suggesting a shift in educational paradigms.

Furthermore, the majority of the studies encompass exploratory approaches (Haluza & Jungwirth, 2023; Yeadon et al., 2023; Victor et al., 2023; Dwivedi et al., 2023; Rozado, 2023; Cascella et al., 2023; Ariyaratne et al., 2023; Yan, 2023; Grünebaum et al., 2023; Kooli, 2023; Jungwirth & Haluza, 2023; Geerling et al., 2023; Gilson et al., 2023; Cooper, 2023; Short & Short, 2023). Similarly, several studies take the form of reviews, such as Ray (2023), Lee (2023b), Thurzo et al. (2023), Carvalho & Ivanov (2023) and Kolides et al. (2023). Moreover, text analysis is employed in studies conducted by Taecharungroj (2023) and Tlili et al. (2023). Also, Farrokhnia et al. (2023) conducts a SWOT analysis, while several other works are opinion, position, technical notes, and commentary essays (Salvagno, Taccone & Gerli, 2023; Lee, 2023b; Halaweh, 2023; Cox & Tzoc, 2023; Karaali, 2023; Rao, 2023; Cotton, Cotton & Shipway, 2023; Pavlik, 2023; Emenike & Emenike, 2023). Finally, Kwon (2023) presents technical notes, and Schöbel et al. (2023) delves into bibliometric analysis concerning research on conversational agents.

The analysis of the literature has shown that there is a notable gap in the existing studies demonstrating the actual impact of ChatGPT on educational outcomes. As AI language models become more pervasive, the broader societal impact on information dissemination, content moderation, and public attitudes needs closer examination. Therefore, it is essential for future research to explore regulations, guidelines, and ethical frameworks to ensure the responsible integration of Gen-AI in various professional fields while addressing both advantages and challenges. Overall, the prior studies presented in Table 1 lay the foundation for in-depth investigations, emphasizing the transformative potential of Gen-AI, ethical considerations, and the need for a well-defined regulatory framework to guide future research and practical applications.

In particular, few studies have proposed theories or frameworks, but notable exceptions include the works of Su & Yang (2023) and Lim et al. (2023). The sooner proposed a theoretical framework called “IDEE” (Identify, Determine, Ensure, Evaluate) for the implementation of educative AI. The framework involves identifying desired educational outcomes, determining the appropriate level of automation, ensuring ethical considerations, and evaluating the effectiveness of using ChatGPT and other Gen-AI in education (Su & Yang, 2023). The latter provides a critical analysis through paradox theory by defining Gen-AI and transformative education, establishing the paradoxes inherent to Gen-AI (Lim et al., 2023). Nevertheless, there is a lack of existing studies that appear to have explicitly focused on creating a comprehensive decision-making tool that can assist stakeholders in all sectors in determining whether Gen-AI should be legislated or restricted. While the studies touch upon various aspects of AI, including its impact on education, ethics, and specific domains like healthcare and journalism, none of them seem to provide a unified framework or tool that guides policymakers, educators, researchers, and professionals across different fields in making informed decisions about Gen-AI usage in education. This notable gap suggests that there is a need for an interdisciplinary effort to develop a systematic and adaptable decision-making tool that can be applied across sectors to address the complex challenges and opportunities presented by Gen-AI effectively. Such a tool could help ensure the responsible and balanced integration of Gen-AI while taking into account the specific needs and concerns of various stakeholders.

Overview of RRR

Roberts (2023) introduced a new framework called risk, reward, and resilience, coined as RRR, which combines knowledge from various disciplines and fields. This framework offers a straightforward yet adaptable mental decision-making model that can be applied to diverse domains. To enhance the chances of survival and success in a complex and uncertain environment, RRR emphasizes the importance of considering the interconnections of three key factors: risk, reward, and resilience. Each of these factors consists of three drivers. Risk is determined by the combination of hazard or threat, exposure, and vulnerability. The reward is influenced by opportunity, access, and capability. Resilience, on the other hand, depends on absorptive, adaptive, and transformative capacities (see Fig. 2). According to Roberts (2023), the risk lies at the point where threat, exposure, and vulnerability intersect. This intersection relies on a combination of the severity of an external hazard or threat and how it interacts with the exposure and vulnerability of a particular entity or system. On the other hand, reward is determined by factors like opportunity, access, and capability. These elements describe the potential benefits that can be gained from a specific action, the circumstances and avenues through which an entity or system can exploit these opportunities, and the internal attributes of the entity or system that influence the gains they are likely to obtain from accessing these opportunities. In the context of RRR, risk primarily concerns negative internal and external aspects (vulnerability and threats), while rewards are associated with positive internal and external factors (capability and opportunity). As a result, capability and vulnerability are used interchangeably with strength and weakness and have broader applicability. RRR framework introduces access as a significant driver for reward, particularly relevant for policymakers whose regulations often impact access.

Furthermore, the core of resilience is found in the ability of entities and systems to absorb, adapt to, and transform in response to ongoing changes (Béné et al., 2012; Roberts, 2023). The increased focus on resilience thinking should be seen as a response to the challenges of operating, conducting business, or governance in an increasingly complex world (Walker & Salt, 2012; Roberts, 2023). The drivers of resilience represent the dynamic capabilities that enable entities and systems to deal with change by absorbing, adapting to, and transforming in the face of threats or hazards (Roberts, 2023). Absorption refers to an entity or system’s capacity to withstand threats or hazards without experiencing significant negative consequences. Adaptation involves the ability of a system to respond to threats or hazards by making adjustments that enable it to continue functioning, albeit in a slightly altered manner. Transformation, on the other hand, pertains to the ability to alter the structures and incentives of the entity or system in a way that not only allows recovery from shocks but also fundamentally changes the entity or system for the future.

The RRR framework incorporates cross-cutting elements that can be applied across multiple domains. Within this framework, risks can be economic, security-related, or health-related. Rewards can take the form of monetary, diplomatic, or social benefits. Resilience can be physical, psychological, or environmental in nature. The RRR framework has the potential to break down silos and promote communication across disciplines, thereby enhancing policy outcomes. It offers a more comprehensive and unbiased approach compared to existing frameworks (Roberts, 2023). By identifying the drivers of each element, understanding their connections, and recognizing the policy choices and consequences associated with them, the RRR framework provides a simplified and structured systems model for tackling complex problems. It explicitly addresses the objectives, trade-offs, and assumptions that underlie policy-making processes. Importantly, the RRR framework does not dictate what individuals or policymakers should think about complex problems. Instead, it serves as a guide to help them understand how to approach such problems. It allows for the inclusion of diverse and sometimes conflicting hypotheses on a single diagram, enabling experts from different fields to see that their insights and values are respected. Simultaneously, it highlights the existence of alternative perspectives that should be evaluated and considered. In this way, the RRR framework facilitates the decision-making process by helping identify the best feasible outcome among various choices.

Research keyword and search process

Research keywords are determined when the systematic review protocol is established, shaping the scope of the retrieved materials (Thornley & Gibb, 2009; Okoli, 2015). These keywords were instrumental in identifying the articles (Ridley, 2012). As a result, relevant research keywords were formulated and employed in the search query to explore articles pertaining to the area of interest in the Scopus database. The selection of research keywords adhered to the guidelines outlined by Okoli (2015), emphasizing the need for researchers to transparently and comprehensively specify their chosen keywords. These keyword terms encompassed phrases like “ChatGPT,” “large language models,” “LLMs,” “ethical issues,” “concerns,” “ethics,” and “implications,” with various combinations such as ChatGPT OR large language models OR LLMs AND “ethical issues” OR concerns OR ethics OR implications; or ChatGPT AND “ethical issues” OR concerns OR ethics OR implications. On 3rd May 2023, a comprehensive search was conducted on Scopus database to gather relevant scholarly articles related to the study’s focus on ChatGPT. The search used the selected keywords, depicted in Fig. 3. The initial search yielded a total of 74 results. To narrow down the search to the most relevant articles, the keywords were specifically limited to ChatGPT, resulting in 47 articles. Subsequently, an Excel file was obtained from Scopus, containing the titles and abstracts of these 47 articles. Each article’s title and abstract were carefully reviewed to ascertain its suitability for inclusion in the study.

Article selection

The initial phase involved the screening of titles and abstracts as the primary criteria to identify and eliminate irrelevant articles. This process assisted the researchers in determining whether the articles aligned with the predefined inclusion or exclusion criteria. The exclusion criteria encompassed non-English content, conference articles, book chapters, and editorial pieces, while the inclusion criteria comprised all articles listed in Scopus. To ensure the data’s quality and relevance, the selection of articles indexed in Scopus ensured that they met the quality assessment standards established for this study. Finally, undertaking the screening process eliminated six articles that were either editorials or not directly aligned with the study’s focus. This step was aimed at maintaining the integrity and coherence of the final dataset (Kitchenham & Charters, 2007; Ahmed et al., 2019). Ultimately, a total of 41 articles were deemed suitable and included as the final study sample. Consequently, these 41 articles were chosen for data extraction, as illustrated in Fig. 3.

Data extraction and analysis

Data extraction served as a means for researchers to obtain specific information from the articles. Thus, the selected articles were subjected to a systematic data extraction process, focusing on the three key terms of the RRR framework: risk, reward, and resilience. Accordingly, data was extracted from the articles according to the aim of the study. The information encompassed themes or topics related to these elements. The objective of data synthesis was to systematically organize and classify the various themes identified in the articles. To achieve this, a thematic analysis approach was employed (Clarke, Braun & Hayfield, 2015; Abedin, Jafarzadeh & Olszak, 2021; Babar, Bunker & Gill, 2018). The study closely scrutinized the information to identify recurring themes, including common topics, ideas, patterns, and approaches. A primary theme was established when the extracted data was found to be related to other sub-themes that shared a similar logical context. For instance, themes like copyright, compliance with copyright laws, consent, and legal issues were grouped together, as well as themes like misinformation, inaccuracy of information, and nonsensical content were grouped together. This process facilitated the researchers in comprehending the diverse themes reported in existing studies. Additionally, the information was assessed and the frequency of terminology was computed to illustrate their occurrence in the literature, which was done through critical analysis as presented in Tables A1–A3 (refer to Appendix A).

RRR application process and guidelines

Policymakers can use the RRR framework to inform their decision-making processes and improve the effectiveness of policies in various domains, such as economics, security, and environmental management. Figure 4 proposes the process of utilizing RRR for decision-making, as illustrated in three (3) steps. According to the framework proposed in this study, policymakers should identify and assess potential risks associated with a specific system. They should be able to analyze threats, vulnerabilities, and exposure related to the system. The risks are weighted and prioritized based on their severity and potential impact. This prioritization can help in determining the most critical issues. Secondly, the potential rewards to be gained from the system are assessed and evaluated, such as the potential benefits and opportunities associated with the system. This is done considering the capability and access factors that can lead to positive outcomes. The rewards are also weighted and prioritized based on their potential positive impact. Thirdly, the potential capability that can make stakeholders withstand the risks associated with the system is identified. In general policy making should be driven by maximizing the benefits and minimizing the risks and using an objective assessment of resilience in weighting the benefits and risks. As a result, our study puts forth a set of decision-making rules, outlined in Table 2, designed to guide and inform the application of the RRR framework in the creation of policies that are both effective and efficient. It is important to acknowledge that these proposed decision-making rules, inspired by Kamali (2015), can be adapted and customized to suit the specific characteristics and demands of the system under assessment for policy development.

Accordingly, the decision-making rules ensure that policies are designed with resilience in mind, focusing on the ability to absorb, adapt, and transform in response to changing conditions. This includes incorporating strategies to mitigate the impact of potential threats, such as promoting adaptability and supporting transformation. Moreover, the guiding rules encourage systems and organizations to adapt and adjust in response to unexpected challenges, making them more robust and capable of withstanding disruptions. In addition, the decision-making rules are designed to facilitate the transformation of people, institutions, or organizations to better address emerging challenges or opportunities. Notably, the RRR framework encourages a dynamic approach to policy-making. The decision-making rules are created to continually monitor the evolving risk landscape, adapt policies as needed, and remain flexible in response to changing circumstances. By applying the RRR framework, policymakers can create more robust and effective policies that not only consider potential risks and rewards but also build resilience in the people or society. This approach can lead to more adaptive and successful policy outcomes in an ever-changing and complex world.

Elements of ChatGPT based on RRR

As a way to help decision-makers make an effective decision regarding LLM tools, this study makes an effort to identify risk, reward, and resiliency aspect associated with LLMs by using the RRR framework (Roberts, 2023), which was identified from various elements of ChatGPT in literature. Accordingly, this information was extracted based on the three themes of the integrative framework. ChatGPT usage was evaluated through the RRR framework to provide a comprehensive report to guide stakeholders and policymakers on the future utilization of ChatGPT, by extension LLMs tools. The proceeding sections discuss these themes and their critical analysis from the literature.

Risk

The list of terms related to “Risks” highlights various challenges and potential negative implications associated with the adoption of generative AI systems. These risks encompass concerns such as privacy, data confidentiality (Dowling & Lucey, 2023; Eggmann et al., 2023; Dwivedi et al., 2023; Rao, 2023; Carvalho & Ivanov, 2023; Masters, 2023; Tlili et al., 2023), output quality (Dowling & Lucey, 2023; Farrokhnia et al., 2023; Su & Yang, 2023), bias (Eggmann et al., 2023; Sallam, 2023; Ray, 2023; Farrokhnia et al., 2023; Masters, 2023; Rozado, 2023; Emenike & Emenike, 2023; Dwivedi et al., 2023), misinformation (Eggmann et al., 2023; Valentín-Bravo et al., 2023; Ariyaratne et al., 2023; Salvagno, Taccone & Gerli, 2023; Carvalho & Ivanov, 2023; Jungwirth & Haluza, 2023; Dwivedi et al., 2023), cybersecurity (Eggmann et al., 2023; Sallam, 2023; Carvalho & Ivanov, 2023), academic integrity (Perkins, 2023; Yan, 2023; Farrokhnia et al., 2023; Cotton, Cotton & Shipway, 2023; Geerling et al., 2023; Emenike & Emenike, 2023; Lim et al., 2023), job evolution (Lund et al., 2023; Taecharungroj, 2023; Carvalho & Ivanov, 2023), copyright compliance (Lund et al., 2023; Sallam, 2023; Lee, 2023b; Masters, 2023; Cooper, 2023), ownership (Lund et al., 2023; Masters, 2023; Victor et al., 2023), transparency (Sallam, 2023; Masters, 2023), reliability (Valentín-Bravo et al., 2023; Thurzo et al., 2023), and ethical considerations. Notably, the deployment of AI systems necessitates paying careful attention to these risks in order to mitigate their potential adverse effects. Privacy concerns emphasize the importance of protecting user data and ensuring compliance with privacy regulations. Data confidentiality is vital to prevent unauthorized access or leakage of sensitive information.

In addition, the quality of AI system output is a significant concern, as inaccuracies, misinformation, and nonsense content can lead to unreliable or misleading results. Bias response poses the risk of perpetuating existing biases, potentially resulting in unfair or discriminatory outcomes. Cybersecurity issues emphasize the need to protect AI systems from malicious attacks or unauthorized access. Academic integrity and honesty concerns encompass risks such as plagiarism (Perkins, 2023; Lund et al., 2023; Sallam, 2023; Qasem, 2023; Salvagno, Taccone & Gerli, 2023; Farrokhnia et al., 2023; Cotton, Cotton & Shipway, 2023; Emenike & Emenike, 2023), incorrect citation practices (Lund et al., 2023; Sallam, 2023; Ariyaratne et al., 2023; Grünebaum et al., 2023; Jungwirth & Haluza, 2023), and lack of originality (Sallam, 2023; Carvalho & Ivanov, 2023). Job evolution and expectations highlight the potential impact of AI on employment and the need to address evolving job roles and expectations. Compliance with copyright laws, consent, and legal issues is crucial to ensure the ethical and legal usage of AI technologies.

Furthermore, the risks associated with ownership, authorship, impersonation, infodemics (Sallam, 2023; Carvalho & Ivanov, 2023; Cooper, 2023; Dwivedi et al., 2023), and lack of deep understanding (Farrokhnia et al., 2023; Grünebaum et al., 2023) emphasize the importance of responsible and accountable AI usage. Declining high-order cognitive and thinking skills (Farrokhnia et al., 2023; Halaweh, 2023; Karaali, 2023) raise concerns about over-reliance (Halaweh, 2023; Qasem, 2023) on AI systems, potentially leading to a decrease in critical thinking and problem-solving abilities. Accordingly, data not being apparently updated (Sallam, 2023; Grünebaum et al., 2023) can result in outdated or irrelevant information being presented by AI systems. Exploitation risks (Kooli, 2023) emphasize the need to prevent the misuse or abuse of AI technologies. Educational equity (Yan, 2023; Emenike & Emenike, 2023) concerns highlight the potential for AI to widen the digital divide if not accessible to all individuals (Salvagno, Taccone & Gerli, 2023; Farrokhnia et al., 2023; Cotton, Cotton & Shipway, 2023; Emenike & Emenike, 2023).

Additionally, safety or security issues from misuse (Yan, 2023; Su & Yang, 2023; Kooli, 2023; Masters, 2023; Cascella et al., 2023; Rozado, 2023; Dwivedi et al., 2023) underscore the importance of implementing measures to prevent AI systems from being used for malicious purposes. Lack of deep understanding emphasizes the limitations of AI systems in comprehending complex or nuanced contexts accurately. Despite the fact that AI systems offer significant benefits, it is essential to acknowledge and address the associated risks. Implementing safety measures, ethical practices, and regulatory frameworks is crucial to ensure responsible AI deployment that respects privacy, fairness, transparency, and accountability while mitigating potential negative consequences. Figure 5 demonstrates the risks and ethics-associated issues related to ChatGPT.

Reward

The list of terms related to “Reward” encompasses a wide range of benefits and positive outcomes associated with various aspects of utilizing AI, particularly in the context of text-related tasks. The key emphasis of these terms is that AI-powered systems provide numerous advantages, such as improved efficiency, effectiveness, accessibility, and support for diverse activities across different domains. Firstly, one prominent theme is the generation of ideas, which includes idea generation (Dowling & Lucey, 2023; Lund et al., 2023; Halaweh, 2023), hypothesis generation (Ray, 2023), and prompt writing (Taecharungroj, 2023). AI systems can assist in sparking creative thinking and generating innovative ideas, contributing to problem-solving and critical thinking skills (Sallam, 2023; Taecharungroj, 2023). Similarly, efficiency and productivity enhancements are highlighted through terms such as data identification (Dowling & Lucey, 2023), streamlining workflows (Sallam, 2023; Yan, 2023; Farrokhnia et al., 2023; Rao, 2023; Carvalho & Ivanov, 2023), cost savings (Sallam, 2023), and increased productivity (Carvalho & Ivanov, 2023; Cascella et al., 2023; Emenike & Emenike, 2023). AI technologies can automate time-consuming tasks, streamline processes, and save resources, enabling individuals and organizations to achieve more performance in less time.

Moreover, enhanced communication and access to information are emphasized through multilingual communication and translation services (Eggmann et al., 2023; Lund et al., 2023), and easy access to information (Perkins, 2023; Cotton, Cotton & Shipway, 2023; Lim et al., 2023). AI systems facilitate effective communication and break down language barriers, enabling users to access and understand information in their preferred language. Also, AI’s support for learning and education is evident in terms like personalized learning (Sallam, 2023; Farrokhnia et al., 2023; Su & Yang, 2023; Cotton, Cotton & Shipway, 2023; Tlili et al., 2023), improved literacy (Sallam, 2023), critical thinking and problem-based solving learning (Sallam, 2023; Taecharungroj, 2023), decreased teaching workload (Farrokhnia et al., 2023; Su & Yang, 2023; Emenike & Emenike, 2023), and teaching and mentoring activities (Emenike & Emenike, 2023). AI-powered tools can adapt to individual learning needs, enhance literacy skills, and assist in creating engaging and interactive learning environments. Hence, text-related tasks, such as text summarization (Eggmann et al., 2023; Jungwirth & Haluza, 2023; Lund et al., 2023), writing fluency and efficiency (Eggmann et al., 2023; Salvagno, Taccone & Gerli, 2023; Yan, 2023), proofreading and editing (Sallam, 2023; Cooper, 2023), text generation (Halaweh, 2023; Cox & Tzoc, 2023; Jungwirth & Haluza, 2023), and assembling or organizing text (Salvagno, Taccone & Gerli, 2023; Jungwirth & Haluza, 2023), benefit individuals and organizations from using AI technologies. These tools can automate and improve the quality of text-related tasks, saving time and effort while maintaining accuracy and coherence.

Additionally, AI’s role in decision support and expertise is highlighted through terms like decision support (Eggmann et al., 2023), supporting expertise and judgment (Kooli, 2023), providing feedback (Su & Yang, 2023; Cotton, Cotton & Shipway, 2023), and supporting professional activities (Emenike & Emenike, 2023). AI systems can provide valuable insights, assist in decision-making processes, and offer expert-level guidance and feedback. Lastly, the extracted keywords also recognize the positive impact of AI on collaboration, transformation, and societal advancements. AI-powered systems foster collaboration and friendship (Cotton, Cotton & Shipway, 2023; Lim et al., 2023), support pitches and presentations (Short & Short, 2023), and align with societal megatrends (Haluza & Jungwirth, 2023), leading to transformation and positive change. Nevertheless, the major highlight of these terms is that AI (as presented in Fig. 6), particularly in the realm of text-related tasks, offers a multitude of rewards and benefits. From idea generation to increased productivity, improved learning experiences, and enhanced decision-making, AI-powered systems contribute to efficiency, effectiveness, and positive outcomes across various domains and activities.

Resilience

The articles reviewed provided various terms associated with resilience, indicating that humanity is showcasing readiness to cope with the challenges of ChatGPT. A typical representation of the resilience elements can be seen in Fig. 7. According to the literature, resilience, as applied to ChatGPT, encompasses a range of considerations that contribute to the model’s robustness, adaptability, ethical usage, and potential for improvement. Similarly, this study explored various terms related to resilience and their implications for the development and deployment of ChatGPT. The extracted terms related to resilience in the context of ChatGPT highlight several keywords. First, it emphasizes the importance of establishing an appropriate testing framework (Dowling & Lucey, 2023; Yan, 2023; Geerling et al., 2023; Thurzo et al., 2023) and acceptable usage guidelines in scientific research (Eggmann et al., 2023; Yan, 2023; Cotton, Cotton & Shipway, 2023; Jungwirth & Haluza, 2023; Ivanov & Soliman, 2023). This ensures that ChatGPT is rigorously evaluated and employed ethically to maintain academic integrity. The concept of co-creation between humans and AI systems emerges as a crucial aspect (Perkins, 2023; Ray, 2023; Halaweh, 2023; Kooli, 2023; Carvalho & Ivanov, 2023; Dwivedi et al., 2023), emphasizing collaboration and continuous feedback loops to refine ChatGPT’s responses and address limitations and biases. Improving human-AI interaction through natural language interfaces and ability features promotes trust and effective collaboration.

Moreover, ethical considerations are central to resilience, and responsible usage is encouraged. This includes solidifying ethical values (Grünebaum et al., 2023; Halaweh, 2023; Kooli, 2023; Cotton, Cotton & Shipway, 2023; Jungwirth & Haluza, 2023; Thurzo et al., 2023; Cooper, 2023; Dwivedi et al., 2023), utilizing AI detector tools (Halaweh, 2023; Cotton, Cotton & Shipway, 2023), and establishing policies and procedures (Rao, 2023; Cotton, Cotton & Shipway, 2023; Carvalho & Ivanov, 2023; Ivanov & Soliman, 2023) to safeguard against potential risks and biases. Raising awareness about the capabilities and limitations of ChatGPT contributes to responsible usage (Kooli, 2023; Cotton, Cotton & Shipway, 2023; Jungwirth & Haluza, 2023). Resilience in ChatGPT extends beyond individual interactions and incorporates considerations of equity and versatility (Sallam, 2023). Efforts to enhance research equity, bridge the digital divide (Ray, 2023), and address diverse language and cultural variations are essential. The aim is to ensure that ChatGPT is accessible and beneficial to a wide range of users. Furthermore, the discussion emphasizes that ChatGPT should not be seen as a replacement for human judgment (Salvagno, Taccone & Gerli, 2023), but rather as a tool to support decision-making processes. Maintaining a balance between AI-assisted innovation and human expertise is vital (Ray, 2023; Cooper, 2023). Furthermore, the study highlights the potential for self-improvement within ChatGPT (Farrokhnia et al., 2023), as well as the need for continuous training, upskilling, and research to enhance its capabilities (Rao, 2023; Cotton, Cotton & Shipway, 2023; Carvalho & Ivanov, 2023; Dwivedi et al., 2023). This ensures that ChatGPT remains adaptable, versatile, and aligned with evolving needs and advancements in the field. Overall, the key highlight from these terms is that resilience in ChatGPT lies in its robustness, adaptability, ethical usage, and the collaborative partnership between humans and AI. By addressing challenges, embracing opportunities, and promoting responsible usage, ChatGPT can effectively augment various domains while aligning with societal values and needs.

Practical application of RRR on ChatGPT ethics conundrum

In the context of ChatGPT ethics, this study aims to adopt the framework (Roberts, 2023) that emphasizes the integration of risk, reward, and resilience for effective policy-making. This framework acknowledges that merely considering risk or reward in isolation is insufficient. Policymakers must internalize both elements and recognize how they are influenced by and impact resilience over time. Understanding the dynamic interplay of these elements is crucial for determining the likelihood of survival and success of actors or systems, such as AI systems like ChatGPT. While risk, reward, and resilience are interconnected, their analysis should not be conducted in isolation. Complex problems, like those related to AI ethics, involve numerous interacting variables. The RRR framework recognizes that changes in one element or driver can have positive or negative effects on others. This interconnectedness can be visualized through RRR diagrams (refer to Figs. 8–10), revealing synergies or trade-offs between risk, reward, and resilience. Additionally, effects can be unpredictable, leading to both positive and negative outcomes.

When it comes to interventions and decision-making, policymakers face choices regarding risk reduction, reward enhancement, and resilience building. The RRR framework encourages a holistic approach, recognizing the equal importance of risk, reward, and resilience rather than prioritizing one over the others categorically. Depending on the context, policymakers may choose to focus on reducing risks, increasing rewards, and/or strengthening resilience, often employing a combination of these strategies. The choice of approach depends not only on the risk tolerance of the actor but also on the nature of the environment. In stable and low-risk environments, maximizing rewards may be the priority. However, in situations with increased risks and uncertainty, actors are likely to prioritize resilience-building to withstand potential shocks. The level of resilience also influences the trade-offs between risk and reward. High risks and rewards may be acceptable if resilience is high, but caution is warranted when resilience is low.

Within the three general intervention categories, policymakers can target specific drivers. For instance, when considering ChatGPT ethics, implementing restrictions on its usage can be a risk-reduction strategy by reducing exposure. Investing in awareness and responsible usage can increase resilience by driving transformative innovation, reducing vulnerability, and enhancing societal absorptive capacity. Generating new rewards can involve accessing new opportunities, such as developing AI detection tools or upskilling, as well as building new capabilities, like detecting and combating misinformation. The RRR framework offers nine potential interventions within these three categories, which can be pursued individually or in combination. Applying the RRR framework necessitates considering the distributional consequences of different interventions across actors, regions, and time. Risks and rewards are often unevenly distributed within a community, and pursuing immediate rewards may come at the cost of increased risks or decreased resilience in the future. Addressing distributional issues is crucial for ensuring fairness, sustainability, and the potential for meaningful change.

In the study, the RRR framework is employed to develop a simplified systems diagram that facilitates an understanding of the connections and policy choices relevant to addressing ChatGPT-related concerns. Figures 8–10 depict causal hypotheses, represented by arrows, indicating whether an increase in one factor leads to an increase or decrease in another. Green arrows denote increasing effects, while red arrows indicate decreasing effects. It is important to note that the existence and magnitude of these effects in practice are empirical questions that can be investigated. The examples and diagrams provided in the study, in accordance with the RRR framework (Roberts, 2023), serve as illustrations rather than comprehensive or empirical representations. They offer roadmaps for exploring empirical inquiries and considering normative trade-offs that policymakers will encounter and need to address.

Summary

When assessing risk, reward, and resilience at an overall educational level, this study discussed the application of the RRR framework to the ethical concerns surrounding ChatGPT. The study explored the interconnected nature of risk, reward, and resilience and how they influence decision-making in various domains. The RRR framework provides a comprehensive and flexible mental model for policymakers to navigate complex problems. Moreover, the study examined the relationship between risk and reward in the context of ChatGPT, acknowledging that certain opportunities presented by the technology can also expose users to risks such as misinformation and copyright issues. Additionally, this study considered how focusing on specific capabilities of ChatGPT can introduce vulnerabilities in other areas, such as educational equity and the digital divide. Similarly, the importance of resilience in addressing the challenges posed by ChatGPT is discussed. While advanced institutions and countries with knowledge acquisition, connectivity, and adaptability have a higher capacity to absorb and respond to these challenges, others may struggle due to outdated teaching paradigms and limited technological resources and readiness.

Furthermore, the role of legislation in managing the impact of ChatGPT was examined within the RRR framework. This study recognized that legislation can have second-order effects, including a potential decrease in rewards by limiting the efficient use of ChatGPT. However, some individuals may be willing to accept these costs if it helps reduce their vulnerability to negative impacts on critical thinking skills. The study also discussed how legislation can enhance absorptive capacity but may simultaneously undermine adaptive capacity in the face of other risks. It was noted that a well-legislated society with responsible ChatGPT usage may exhibit less interest in absorbing or adapting to additional responsibilities. Hence, this study highlighted the importance of considering the interconnected nature of risk, reward, and resilience when addressing the ethical concerns associated with ChatGPT. The RRR framework provides policymakers with a valuable tool to navigate complex decision-making processes and consider the potential second-order effects of various policy interventions.

Key outcome

The application of the framework discusses the interconnected nature of risk and reward in the context of ChatGPT within higher education. The study highlighted that access to ChatGPT offers opportunities for efficiency, like text summarization and workload reduction, but also introduces risks such as plagiarism and cheating. Moreover, using ChatGPT to access vast information can yield rewards but bring risks like misinformation and copyright issues. Developing tools to detect plagiarism and misinformation can enhance academic integrity but may create vulnerabilities in areas like the digital divide and job losses. This risk-reward balance explains why policymakers hold opposing views on ChatGPT, some emphasizing its rewards and others worrying about the risks and neglect of critical thinking in education. Resilience can be increased in some aspects through knowledge and connectivity but can also pose challenges in others. In addition, the study also highlights the choices higher education institutions face in response to ChatGPT challenges, including considering legislation and restrictions. Legislation’s second-order effects can be both positive and negative, potentially decreasing rewards due to limitations but also reducing vulnerability to negative effects on critical thinking skills. Legislation can enhance absorptive capacity but may also affect resilience in various aspects, depending on its efficiency and societal attitudes towards responsibility.

Accordingly, through the application of Gen-AI to the context of higher education, this study demonstrates the RRR framework’s efficacy in addressing the concerns related to ChatGPT. It emphasizes the necessity of understanding the intertwined relationships between risk factors such as bias and misinformation, rewards including enhanced educational experiences, and resilience considerations like adaptability to emerging challenges. As AI technologies evolve, it becomes evident that effective policy responses must transcend simplistic solutions and instead embrace a holistic understanding of the intricate interactions between AI, society, and education. This study underscores the urgency of fostering collaboration among stakeholders—policymakers, educators, AI developers, and the public, to develop comprehensive and forward-looking policies that prioritize responsible and inclusive ChatGPT deployment. Therefore, in the pursuit of cultivating an environment where Gen-AI tools are harnessed to empower rather than undermine human progress, this research offers invaluable insights and guidance. As we navigate the ethical conundrums posed by ChatGPT, this study’s findings shed light on the ethical imperatives that guide our interactions with AI. Through a balanced consideration of risks, rewards, and resilience, we can pave the way for a future in which AI technologies enhance the fabric of higher education while preserving the foundations of a resilient and thriving society. Also, by embracing the lessons derived from this research, policymakers and higher education institutions can chart a course toward an AI-augmented future that is marked by responsibility, inclusivity, and collective wisdom.

Limitation and future research

This study is not without limitations, opening an avenue for future studies. Firstly, the primary focus of our article is the development and application of the RRR framework within the context of the ChatGPT ethical conundrum. The literature review served as a foundation to identify and extract the key elements necessary to practically demonstrate the dynamics of the RRR framework in a specific context, in this case, ChatGPT in higher education. This study assumed that the elements identified from the literature are relevant to the core elements of the framework (risk, reward, and resilience) to provide a clear, context-specific foundation for our research. However, the literature search was conducted in May 2023 and was limited to the Scopus database. This could present concern about the breadth of the literature review which is considered as a limitation of the current work. Therefore, future research could expand the literature search to encompass a broader range of sources. This can certainly explore the incorporation of additional, diverse sources in the literature review to offer a more comprehensive understanding of the topic and to further validate the application of the RRR framework. By doing so, future research could enhance the framework’s robustness and applicability across a wider range of scenarios.

Secondly, this study is subjective and conceptual and conducted in an effort to contribute to the theory regarding ChatGPT decision-making. Subjective research relies on the interpretation and analysis of observations, which can be difficult to quantify and measure objectively. Additionally, the researcher’s own biases can influence the results of the study. After populating the key elements into the RRR framework. The analysis of the ChatGPT ethics conundrum was done based on qualitative reasoning to demonstrate the applicability of the RRR framework to help stakeholders and decision-makers make an informed decision regarding chatGPT by evaluating its risk, reward, and resilience alongside each other. In the future, the RRR elements can be analyzed objectively through a robust and scientifically proven multi-criterion decision-making framework. In particular, future studies can utilize the AHP to formulate the problem into the decision matrix. This will help stakeholders and policy makers make transparent, efficient, and effective decisions and avoid biases frequently associated with qualitative reasoning. The output of the AHP can provide weight to the sub-elements under the three key classifications (risk, reward, and resilience). The decision can be reached by analyzing the weighting obtained by the analysis.

In addition, this study opened up a new direction for researchers to propose ethical frameworks and guiding rules for decision-making for RRR and other systems generally. The development of other ethically balanced rules as an extension can be explored in the future. The decision-making rules proposed in this study are only guidelines and there are bound to be exemptions in specific circumstances based on the certainty on whether rewards can truly be accrued, risks can be truly averted, or resilience can be achieved. It is essential to acknowledge these limitations in the present research. As a result, future research should aim to address this gap by devising strategies to mitigate the potential limitations. This could involve exploring alternative databases, methodologies, case studies, or theories.

Critical analysis of risk elements

Refer to Table A1.

Critical analysis of reward elements

Refer to Table A2.

Critical analysis of resilience elements

Refer to Table A3.