Data ethics is a critical aspect of responsible AI focusing on the ethical aspects of data operations, including data collection, processing, and use. It encompasses a broad range of ethical issues related to data handling, whether the data are structured or unstructured, across various modalities [15-17]. Existing research indicates that poor data ethics can have significant consequences. For instance, 57% of consumers reported they would stop doing business with companies that breached their trust through irresponsible data practices [15]. Moreover, using suboptimal data to train machine learning algorithms can result in harmful outcomes for patients, such as misdiagnosis [18], misidentification [18,19], or data privacy breaches, even if the data were deidentified before use in machine learning algorithms [8].

In the context of this viewpoint report, it is crucial to analyze data independently from the machine learning algorithms that process it. This separation underscores the need for strategies, governance, a data-literate organizational culture, and ethical principles that ensure the creation and use of high-quality data for AI technologies. As Radziwill [20] noted, human error is often the root cause of bad data. Data may also reflect historical human biases embedded during their production, consumption, curation, and management. Therefore, data ethics is essential for addressing issues and unintended consequences stemming from poorly managed data.

Furthermore, there is a growing recognition of the importance of distinguishing data ethics from AI ethics, allowing for ethical decisions, specifically in data handling and the AI algorithms that use this data [16,17,21]. This approach emphasizes a data-centric perspective that focuses on the moral dimensions of data [16] and the ethical principles governing real-world and synthetic data [22].

Generative and predictive machine learning algorithms require extensive data sets to achieve performance objectives. Data scarcity has emerged as a significant concern within the AI and research communities, primarily due to the depletion of real-world training and validation data sets necessary for AI development. Several factors contribute to this issue, such as website owners increasingly protecting their data by enforcing data consent requirements [23]. It is projected that data scarcity could become a critical issue between 2026 and 2032 [24]. Additionally, the use of real-world data for AI development encounters problems such as missing data, which demands either imputation or deletion.

The generation of synthetic data is being accelerated to address issues related to data scarcity, privacy, and consent [22]. Synthetic data are expected to surpass real-world data by 2030 [22,25]. In the health care sector, synthetic data are used for simulation and prediction research, health IT development, education, and training [22,26]. Despite its benefits, generating synthetic data for AI development presents ethical dilemmas and risks, especially in high-stakes areas such as health care. For instance, although synthetic data can help represent diverse populations and reduce algorithmic biases, overreliance on such data can lead to challenges and unforeseen long-term effects of converting unrepresentative data into representative data [22]. Thus, ethical concerns arise regarding the non-maleficence and fidelity of synthetic data—whether it can address real-world disparities or prevent the dissemination of misinformation [22].

Rhem [8] identified eight principles of AI data ethics, which are summarized as (1) transparency: is there clarity regarding the use, purpose, storage, and protection of the collected data? (2) Fairness: does the data collection and usage avoid exacerbating existing inequalities or biases? (3) Privacy: does the data collection process respect individuals’ privacy and autonomy, potentially through informed consent? (4) Responsibility: are data collectors and users accountable for ethical data collection and usage, including any harm resulting from these processes? Are mitigation steps in place? (5) Security: are data stored and transmitted securely to prevent unauthorized access, use, or disclosure? (6) Inclusivity: does the data collection and usage process ensure the inclusion of diverse perspectives and experiences, especially those that are underrepresented? (7) Transparency in decision-making: are decisions based on explainable and interpretable data? (8) Continual assessment: does the organization continuously monitor and assess its data practices to ensure they align with ethical principles [8]?

Shanley et al [22] were inspired by the five principles of responsible AI to initiate discussions on data ethics and synthetic data in AI. They proposed five principles to govern synthetic data (1) responsibility; (2) non-maleficence; (3) privacy; (4) transparency; and (5) justice, fairness, and equity [22]. These principles closely align with the globally recognized principles of AI—responsibility, non-maleficence, privacy, transparency, and justice and fairness [27]. Moreover, the ethical principles suggested by Shanley et al [22] and their associated questions correspond with Rhem’s [8] AI data ethical principles. This alignment is detailed in Table 1, which includes an adaptation of Rhem’s [8] principle of fairness.

Patient safety is paramount in health care and nursing, encompassing the protection of patients from events such as danger, risk, or injury. Security concerns freedom from danger or threats, whereas privacy analyzes the unauthorized access or use of patient data. We suggest incorporating a ninth principle, safety, into Rhem’s [8] 8 principles of AI data ethics, as outlined in Table 1. A pertinent question for this new principle is “Do data operation activities and processes consider, identify, and mitigate risks associated with preventing danger, risk, or injury to individuals (patients)?” There is a dearth of consensus in the scientific and industrial literature from AI and health informatics research communities regarding definitions and practices that would guide AI data ethics. These definitions and shared principles are crucial for guiding the implementation and assessment of data ethics in clinical practice. As noted by Panai [21], data ethics represents an underdeveloped area within organizations and lacks a clear definition. Similarly, AI data ethics is a latent or underexplored area in health care and nursing informatics scientific research.

Numerous proposals exist for ethical data frameworks that support these principles. Floridi and Taddeo [16] advocate for a macroethical data ethics framework, which aims to avoid narrow, ad-hoc approaches and enables organizations to develop solutions that optimize the societal benefits of data science. Furthermore, Marcovitch and Rancourt [17] endorse standardized tools that facilitate the integration of data ethics accountability mechanisms, such as disclosure and transparency processes, at the organizational level. This is particularly important given the variations in legal frameworks across countries. Their proposed tools include the integration of data ethics into organizational culture, the establishment of data processes or management systems, a data governance structure, organizational transparency in ethical decision-making regarding the data supply chain, and a consistent method for demonstrating and verifying ethical data practices [17].

Note that the frameworks proposed by Floridi and Taddeo [16] and Marcovitch and Rancourt [17] are not specifically tailored to the context of health care or nursing. The absence of a verifiable theoretical or conceptual AI data ethics framework that supports research, alongside a practical evaluation framework that translates effectively into practice, represents a significant gap in health and nursing informatics research. The ideas presented by Floridi and Taddeo [16] and Marcovitch and Rancourt [17] could serve as a foundation for developing such frameworks within the nursing field.

Without a shared organizational data vision, strategy, and policies, implementing aspects of the proposed data ethical frameworks may be challenging, including risking data integrity. There needs to be clarity and knowledge about the responsibilities and liabilities of the people in charge of the data processes. The accountability and culture change should begin with the organization’s executive team. Then, the executive team members intelligibly communicate the data cultural expectations, and policy changes to their departments, units, and teams. There is clarity about the responsibilities and liabilities of people who produce and consume the organizational data assets.

Nurses are producers and consumers of organizational data assets. For instance, they produce electronic health record (EHR) data and use them for nursing quality improvement initiatives. Positive deviance in effective data practices could be the impetus for remarkable cultural changes when poor data management practices are in place due to the absence of an organizational-level data vision, strategy, and policies. Suppose a nurse leader is passionate about the societal benefits of data science and AI, including taking the initiative to understand the importance of data quality. This individual becomes a change agent for their team. Their team’s culture becomes data-centric, with patient safety and outcomes at the forefront. The team develops policies and standardized procedures that facilitate improved EHR data entry processes, which are less burdensome for the nurse but help them be accountable to the principles of AI data ethics. The team expresses knowledge about how data quality impacts patient care and the technologies they use in clinical settings. They observe technology as a mechanism to provide quality nursing care. Their attitudes and behaviors result in improved data quality for the nursing unit. This team is now a data vanguard. Other departments and nursing teams notice that this nursing unit outputs high-quality data, leading to improved insights for the unit. This “positive deviant” team’s data practices led to informative unit-level reports, fewer data-related errors, and enhanced patient outcomes. Other nursing teams are curious and want to model this team’s effective data procedures and practices. This influence could motivate the organization’s nursing departments to follow suit.

AI technologies that leverage machine learning require substantial data for effective functioning. Machine learning is a subset of AI that enables computers to learn and adapt autonomously through algorithms and statistical models, with minimal or no human intervention. These systems demand extensive data volumes, with generative AI requiring even larger data sets and greater computing power to discern underlying patterns in the data. Historically, the development of prevalent AI technologies has adopted a model-centric strategy, prioritizing the machine learning algorithm or model as the primary focus for enhancing performance [28]. In this approach, data-related activities such as curation, collection, and labeling are often deprioritized and occur only once, leading to potential ethical issues such as algorithmic bias and mispredictions.

The model-centric approach does not adequately address the principles of data ethics, as it overlooks the complexity, nuances, challenges, and accuracy of data, which are vital for improving the behavior of machine learning models [29]. Recognizing these limitations, the AI industry is shifting toward a data-centric strategy, which places data—the “fuel” of AI—at the core of the development process [28-30]. This strategy emphasizes the importance of data quality to achieve high-performance machine learning models [9,29]. Unlike the model-centric approach, data activities in the data-centric strategy are iterative, while model optimization remains static [9,29]. This approach also promotes the involvement of domain experts to secure relevant, high-quality data sets for machine learning [30].

The data-centric AI methodology involves developing, iterating, tracking, and maintaining the quality and integrity of AI systems’ data. It focuses on creating adequate training data, designing appropriate inference data, and ensuring data sustainability (establishing data lineage) [29]. Given the importance of high-quality data, the data-centric approach aligns with the principles of data ethics and responsible AI. For instance, meticulous data curation and collection support ethical principles of fairness, responsibility, and transparency. The involvement of domain experts such as clinicians, clinical informaticists, and regulatory and privacy specialists promotes inclusivity, transparent decision-making, and privacy protection. These diverse perspectives help safeguard patient data privacy and prevent harm from substandard data. Considering data sets as a distinct and valuable product separate from the machine learning model also encourages health care organizations to establish continuous assessment protocols for their data collection and usage practices.

As AI becomes increasingly integral to the daily responsibilities of nurses across various clinical and administrative settings, nurse leadership must promote a data-centric culture within the nursing profession. Leaders should serve as role models, emphasizing the importance of data quality in nursing practice. One approach to achieving this is by enhancing nurses’ understanding of AI and data. Nurses need to comprehend how downstream systems use the data they produce in EHR. It is crucial to reflect, correct, and evolve from the existing cultural norms and power dynamics that hinder effective EHR data collection, establishing a new culture that recognizes the importance of maintaining data quality at the point of care.

Additionally, nurse educators and researchers should strive to improve their AI and data literacy skills and develop new curricula to bridge the knowledge gap in AI and data among nursing students, particularly those enrolled in nurse informatics programs or those pursuing research in informatics. Data literacy is the ability to explore, read, write, understand, and communicate meaningfully within a specific context [39,40].

Improved data literacy facilitates discussions regarding data usage and ethics in AI. Establishing a common language that clearly defines data literacy and ethics in nursing is necessary, including the principles of AI data ethics to guide the development of nurse-specific AI technologies. Furthermore, a shared language and a solid foundation in data knowledge equip nurses to act as data stewards, engage in ethical discussions, innovate in data management, and collaborate with AI specialists to develop responsible, nurse-specific AI technologies. These initiatives align with the principles of inclusivity and responsibility in AI data ethics.

Creating and nurturing a data-oriented culture among nurses can enhance their involvement in ethical accountability mechanisms for AI data, as outlined by Marcovitch and Rancourt [17]. However, the concept of data ethics remains poorly defined, leading to overlapping responsibilities and ambiguous accountabilities. This lack of clarity can hinder the effective detection of ethical violations in data use, as data ethics often merges into broader ethical principles that are insufficient at the granular level of data abstraction [21]. Consequently, there is a need for specific job roles focused on how data ethics can inform ethical decisions regarding data and the machine learning algorithms that process it. These roles would involve establishing relevant policies and practices and advocating for the consumer—here, the patient. Thus, the recommendation to establish a Chief Data Ethics Officer role emerges [21,41]. The mandate of this officer is ethical rather than legal [41], focusing on leading a support team responsible for drafting a code of data ethics, forming and managing a data ethics committee, and overseeing data-oriented ethical issues [21].

With a data-centric organizational culture, data governance, and a Chief Data Ethics Officer, nurse leaders and researchers can explore and implement innovative, nursing-centric data roles. These roles are crucial for upholding the principles of AI data ethics in nursing practice and creating standardized data accountability tools, policies, and processes that effectively evaluate and measure AI data ethics in nurse-specific AI technologies. Job roles such as Chief Nurse Data Ethics Officer and Nurse Data Steward should be explored further.

Nurse educators should implement courses that prepare nurses for data-oriented roles in AI. Nursing degree programs at both undergraduate and graduate levels should be adapted to include relevant courses on AI, data ethics, foundational data science, and data literacy. Nurse scientists are encouraged to explore novel approaches, frameworks, and instruments that enable the integration of ethical accountability mechanisms and effectively assess the principles of AI data ethics reinforced by scientific evidence.

Domain expertise is essential for generating high-quality data [32], and human involvement is critical to successfully executing data-centric AI tasks [29]. Nurses can use their knowledge of data-centric AI and data ethics to actively engage in all phases of AI development and positively leverage their expertise. Specifically, nurse informaticists have opportunities to participate in activities such as data annotation, labeling, and the verification and validation of data elements used in AI data sets. Furthermore, throughout the AI development lifecycle, nurses can play a pivotal role in identifying and addressing opaque data decisions that may affect frontline clinicians’ trust and usage of AI technologies, thereby enhancing transparency in decision-making processes. The active involvement of nurses in these data-centric AI activities, combined with their commitment to patient advocacy, supports the ethical principles of data privacy and promotes accountability and ownership of data collection, processing, and use.

Nurse leaders, educators, and researchers must recognize that implementing responsible AI technologies and initiatives in health care is complex and challenging, despite significant interest in AI ethics [5]. This statement is not intended to deter nurses from exploring methods to develop ethical AI technologies that use high-quality, representative clinical data. Instead, it aims to raise awareness and encourage nurses to persevere through challenges, including maintaining patience and resilience during the change management process. AI technologies are often politically influenced, reflecting their designer’s values, beliefs, and norms, as well as the data and data sets used to train them [5,42]. Political consequences may arise from data operations such as selection, labeling, preprocessing, and transformation [42]. Nurses, working at various touchpoints across diverse settings in the health care ecosystem, bring unique perspectives to the AI discussion, which can help promote and advocate for the principles of AI data ethics.

Working with data is time-consuming and challenging and often perceived as less exciting than developing machine learning models or AI technologies. Despite this, the foundational role of data in AI systems cannot be overstated; high-quality data are crucial for the performance and value of AI technologies in health care. The principles of AI data ethics aim to promote responsible AI and the creation of ethical AI technologies [8]. The methods used to collect, store, use, and share data have profound implications for individuals, organizations, and society [8].