This narrative review aims to assess the benefits and risks associated with the integration of AI into health care, with a primary focus on potential biases, transparency issues, data privacy concerns, and safety risks. A literature review was conducted to explore the current landscape of AI applications in health care and to identify relevant ethical, regulatory, and safety considerations.

Specific inclusion and exclusion criteria were established to guide the selection of studies for this narrative review. Studies were included if they were relevant to the 3 core concepts of AI, ethics, and health and were written in the English language. Articles were excluded if they did not explicitly address each of the core concepts of AI, ethics, and health or if they were not written in English. In addition, studies focusing solely on ethics and big data without explicit mention of AI methods or applications were excluded. Non–peer-reviewed academic literature, such as letters and non–peer-reviewed conference proceedings, as well as books and book chapters, were also excluded as they were deemed irrelevant to this review. No restrictions were applied regarding the publication date or study design to ensure a broad overview of the topic.

The literature search used 8 electronic databases: OVID MEDLINE (1946-present), OVID Embase (1947-present), OVID PsycINFO (1806-present), EBSCO CINAHL Plus with Full Text (1937-present), ProQuest Sociological Abstracts (1952-present), ProQuest Philosopher’s Index (1940-present), ProQuest Advanced Technologies & Aerospace (1962-present), and Wiley Cochrane Library. Search strategies were tailored to each database (Multimedia Appendix 1), using controlled vocabulary, Medical Subject Headings (MeSH) terms, EMTREE terms, American Psychological Association’s Thesaurus of Psychological Index Terms, CINAHL headings, Sociological Thesaurus, Philosopher’s Index subject headings, and Advanced Technologies & Aerospace subject headings. The searches were limited to English language–only articles, and filters excluding animal studies were applied to specific databases. In addition, a filter for health or medicine-related studies was applied to the Advanced Technologies & Aerospace database.

The final searches of the peer-reviewed literature were completed on June 23, 2023. Gray literature was not searched in this narrative review.

All identified records from the academic literature searches were imported into the reference management software EndNote (Clarivate). After removing duplicate records, screening was conducted in 2 steps: initial title and abstract screening followed by full-text review. Full-text reviews were conducted to ensure that the selected studies provided substantial insights for the narrative synthesis.

Data charting forms were developed and refined based on the narrative review research question. The forms included fields for recording data such as the objective of each paper, institutional affiliations of authors, publication year, country of the first and corresponding authors, conflict of interest disclosures, health context of interest, AI applications or technologies discussed, ethical concepts, issues or implications raised, reference to global health, and recommendations for future research, policy, or practice. Data were recorded directly into the data charting form with corresponding page numbers to ensure accuracy.

Data analysis included thematic components. Thematic analysis was conducted inductively, generating open descriptive codes from a sample of records. Codes were applied to relevant data points across all records, with new codes added as needed. These codes were then organized into themes, allowing for the identification of commonalities and gaps in the literature. Results are presented in a narrative format.

Within the realm of integrating AI into health care, this narrative review has revealed a broad range of insights that span a spectrum of possibilities and challenges. This section categorizes the findings into 2 overarching categories: “Benefits” and “Risks.” Each category encapsulates a tapestry of themes that emerged from an exploration of academic literature. As these themes are explored, the multifaceted landscape of AI’s influence on health care is illuminated. The “Benefits” section unveils the potential for AI to revolutionize health care delivery, ushering in more accurate diagnoses, personalized treatment regimens, and streamlined resource allocation. Conversely, the “Risks” section delves into the intricate ethical, privacy, and safety concerns that accompany the integration of AI into clinical settings. Through a comprehensive examination of these themes, this review provides a nuanced perspective on the implications and imperatives in harnessing AI’s potential for the betterment of health care.

The systematic literature review process, as illustrated in the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) flow diagram (Figure 1), outlines a thorough and rigorous methodology. Initial searches across multiple databases—MEDLINE, Embase, PsycINFO, ProQuest Sociological Abstracts, CINAHL, ProQuest Philosopher’s Index, ProQuest Advanced Technologies & Aerospace, and Cochrane Library—yielded a total of 8796 articles. After removing 4798 duplicates using Endnote, 3738 unique records were screened for relevance. Of these, 3155 articles were excluded based on title and abstract review for not meeting the inclusion criteria. The remaining 583 articles underwent full-text assessment for eligibility. Further evaluation led to the exclusion of 539 articles due to various reasons, such as unavailability of full text (n=171), irrelevance to the primary outcome (n=290), non-English language (n=22), not being peer-reviewed (n=29), and not being original research (n=27). Ultimately, 44 studies were included in the qualitative synthesis and data extraction. This meticulous selection process ensured that the final set of studies provided a robust and representative foundation for examining the integration of AI in health care.

Textbox 1 below describes the main benefits of implementing AI in health care. The benefits are explained in detail below.

AI adoption in health care offers various medical, economic, and social benefits. This section discusses some of the key medical benefits of AI.

In addition to the medical health benefits, using AI in health care has other economic and social advantages, as discussed below.

The risks of AI in health care are listed in Textbox 2.

This narrative review delves into the dynamic landscape of AI integration in health care, aiming to uncover a spectrum of perspectives, concerns, and opportunities. This exploration encompasses a diverse range of health care settings from different countries and regions, unveiling a rich tapestry of AI adoption. Overall, AI offers tremendous potential and will continue to play a crucial role in future health care decisions. If AI is successfully used, it can reduce pressure on health care workers while improving work quality by lowering mistakes and improving precision. It has the potential to give people more control over their health decisions and can lower avoidable hospitalizations. It can broaden the scope of medical knowledge and build on the present clinical guidelines. Given its advantages and capability to drive the development of precision medicine, it is universally acknowledged to be a much-needed enhancement in medicine. AI is anticipated to eventually master most of the essential domains within health care.

However, there are some difficulties associated with incorporating AI in health care. Acquiring enough data to train precise algorithms is a continual effort that necessitates a shift in attitude towards data sharing that promotes technical advancement. Specific guidelines on how to securely adopt and evaluate AI technology and research on AI’s potential and limitations are required. Robust research is also required to empirically demonstrate the benefits of AI use in the actual world. While the perfect conditions for successful AI adoption may not yet exist, there is still room for AI advancement in health care.

Given that AI has tremendous potential and is the future, there are a few crucial points to consider when using AI in health care. First, given the need for more general agreement in AI governance, it may be impossible to develop AI-based systems whose algorithms can be generalized across all health care domains. As a result, it may be wise to concentrate on systems that can be implemented and used effectively in the health care institutions for which they were designed. Fundamentally, patient care must take priority over the excitement of cutting-edge technology. The AI system’s safety and competence must be weighed for use only when appropriate and valuable to patients.

Second, AI in health care must still be complemented with human input. Although AI has advantages in speed and accuracy, physicians are still needed for more cognitively complex or psychological elements and activities. Similarly, although the detection and monitoring of vital disease symptoms are now automated, the objective behind AI is not to eliminate physician input but to focus their expertise on areas where they are more necessary and on what computers cannot and may never imitate. Therefore, focusing on developing complementarity between the use of AI and physicians by training is essential.

In addition, while it is critical to lower expectations, it is also critical not to be excessively gloomy about the role of AI in health care. While physicians may need to comprehend the processes of AI algorithms, most physicians understand magnetic resonance imaging or computed tomography to some level. Despite a lack of individual physician comprehension of their specific process, these studies are extensively used. The lack of transparency in ML algorithms may thus be tolerable if the algorithm’s efficacy can be demonstrated. Again, this can be achieved with training and familiarizing the physician with the AI system.

Rather than putting AI to a standard of either perfect or nil results, one should compare the outcomes of using AI to those of the natural world, where physicians can and will make mistakes. Importantly, AI is dynamic in nature and can improve using more extensive data sets. As a result, it is entirely possible that the combined usage of physician and AI input would be more successful over time. However, it is critical not to overstate the status of AI. Its implementation in health care will be a careful, deliberate, and progressive process, including strict control and monitoring of its use and efficacy. AI can help patients and increase the quality of care when combined with input and oversight from health care professionals. AI systems will not wholly replace human clinicians but will supplement their efforts to care for patients. Human therapists may eventually shift towards jobs and job designs that require distinctly human skills, such as enthusiasm and knowledge to use AI in health care.

As global communities live longer lives and the prevalence of chronic disease rises, the rising cost of health care will remain a hot topic among health care stakeholders. It is time to seek the assistance of machines as they can potentially reduce economic costs. Furthermore, coordination between government and private sector industry partners is vital to realize this potential and take advantage of AI’s full potential in health care delivery.

With all this, the key challenge for AI in many sectors of health care is ensuring its adoption in daily clinical practice rather than whether the technologies will be equipped to be effective. AI systems must be approved by regulators, linked with electronic health record systems, standardized to the point that similar products perform similarly, taught to medical practitioners, paid for by public or private payer organizations, and modified in the field over time for widespread adoption to occur shortly. Since AI has a significant and lasting impact on lives and is the future of health care, it is essential to address the associated concerns. Given its importance, AI needs proper policy guidelines and regulations regarding its usage in health care to reap its maximum benefits.

In comparing the findings of this review with existing literature, several key similarities and differences emerge. This review aligns with Gazquez-Garcia et al [63] and Mooghali et al [64] in highlighting the crucial role of health care professional training for effective AI integration. Both emphasize the need for proficiency in AI fundamentals, data analytics, and ethical considerations, reinforcing the notion that successful AI adoption requires a well-prepared workforce. The review also echoes Sapci and Sapci’s [65] advocacy for incorporating AI education into medical curricula to address future challenges.

However, this review diverges in its emphasis on practical AI implementation challenges. While Moghadasi et al [66] and Muley et al [67] discuss the risks associated with AI, including the need for enhanced transparency and stakeholder collaboration, this review adds a nuanced perspective on balancing AI’s potential benefits with its ethical risks. For instance, this review highlights the importance of human oversight and the complementarity of AI with clinician expertise, which aligns with Morley et al [68] and Zhang and Zhang [69] but also offers additional insights into practical implementation issues not fully covered in the previous reviews.

In terms of public perception, this review supports Kerstan [70] and Castagno and Khalifa [71] by acknowledging that trust in AI is influenced by knowledge and transparency. However, it further explores the dynamic interaction between AI’s promise and the necessity for rigorous validation and ethical governance, as discussed by Macrae [72] and Tulk Jesso et al [73]. This review underscores that while AI has the potential to revolutionize health care, its integration must be handled with careful consideration of both practical and ethical dimensions to achieve meaningful improvements in patient care and outcomes.

First, the generalizability of the findings may be affected by the inherent variations in study methodologies, AI implementations, and health care settings across different regions. This heterogeneity introduces variability that could influence the applicability of the conclusions. To mitigate this limitation, rigorous search strategies were used across multiple databases to include a diverse range of studies. Future reviews could benefit from incorporating more standardized studies to enhance generalizability. Second, the reliance on published literature from electronic databases introduces potential publication bias. Studies with positive outcomes related to AI integration in health care may be more likely to be published, which could skew perceptions of AI effectiveness and adoption rates. Efforts were made to address this bias by including a broad range of databases and emphasizing recent literature. Future research should aim to include unpublished studies or grey literature to provide a more balanced view. Third, the rapid evolution of AI technologies means that newer developments and implementations may not have been fully captured in this review. The review focused on the most current literature available at the time of the search to address this issue. Regular updates will be necessary to incorporate the latest advancements and ensure the review remains relevant. In addition, the absence of details around stakeholder engagement could have enriched the study by providing additional depth and perspective. Engaging stakeholders in such a dynamic field would offer diverse viewpoints and further validate the conclusions. Future research should consider incorporating stakeholder engagement to enhance the robustness and applicability of the findings.

Despite these limitations, this review offers several notable strengths. It provides a comprehensive overview of AI integration in health care, leveraging rigorous search strategies across multiple databases to ensure a diverse and current collection of literature. This approach contributes to a nuanced understanding of AI’s potential and limitations. Furthermore, the emphasis on recent developments helps ensure that the review reflects the most current trends and advancements in AI technologies.

Moving forward, further research in the field of AI integration in health care should address several key areas to advance understanding and application. First, studies should prioritize incorporating stakeholder engagement, including health care providers, patients, policymakers, and technology developers, to provide diverse perspectives on AI adoption and implementation strategies, enhancing relevance and acceptance in clinical practice. Second, longitudinal studies are crucial to assess the long-term impacts of AI technologies in health care settings, providing insights into sustainability, scalability, and real-world effectiveness over time. Third, comprehensive research focusing on the ethical implications of AI, including data privacy, algorithm bias, patient consent, and regulatory frameworks, is needed to build trust and ensure responsible deployment. In addition, comparative effectiveness research comparing AI-assisted interventions with standard care protocols can provide evidence of AI’s impact on clinical outcomes, patient safety, and health care efficiency. Interdisciplinary collaboration between computer scientists, health care professionals, social scientists, and ethicists is essential to foster innovative approaches aligned with health care needs. Education and training programs for health care professionals on AI technologies will ensure proficiency in interpreting AI-generated insights and integrating them into patient care effectively. Finally, research should explore how AI can reduce health care disparities and improve access to quality care, particularly in underserved communities and low-resource settings. Addressing these priorities will realize AI’s potential in transforming health care delivery and improving patient outcomes globally.

In summary, AI presents a transformative force in health care, with the potential to enhance patient care, reduce errors, and broaden medical knowledge. However, its successful integration requires adaptability; complementarity with human expertise; transparency; and a deliberate, incremental approach. AI’s impact on health care is evolutionary, not revolutionary, and collaboration between stakeholders, standardization, education, and robust policies are essential to harness its full potential while upholding patient-centric care and innovation.