Differences in Cultural Views on AI in Healthcare: A Call for Local Engagement

1. Introduction

Artificial Intelligence (AI) has emerged as a transformative force in healthcare, promising more efficient, personalized, and predictive care [1]. By leveraging large-scale datasets and advanced algorithms, AI-driven technologies can interpret complex medical information, improving diagnostic accuracy, optimizing treatment plans, and enhancing clinical workflows [1]. These advancements have already demonstrated tangible benefits in domains such as radiology, cardiology, and oncology, where AI models have achieved performance comparable to or surpassing human experts. However, despite this rapid technological progress, the real-world adoption of AI in healthcare remains limited, particularly outside high-resource environments [2].

A growing body of literature identifies social, ethical, and cultural barriers as major determinants of AI’s limited implementation [3,4,5]. Most AI-driven technologies are trained on data from Western and high-income populations, leading to performance gaps and trust issues when deployed in different cultural contexts [6,7]. Furthermore, studies show that patient perceptions of AI in healthcare are deeply shaped by local beliefs, values, and social norms [8]. For example, attitudes toward data sharing, trust in automation, and the acceptability of algorithmic decision-making differ widely between Western, Eastern, and Global South societies [9]. These variations influence not only patient acceptance but also clinician trust, regulatory readiness, and ethical oversight of AI-driven healthcare.

Despite increasing recognition of these socio-cultural factors, the literature on AI ethics in healthcare remains largely theoretical or focused on technical fixes for algorithmic bias [10,11]. A critical, empirical gap persists: a lack of in-depth, multi-stakeholder case studies that move from identifying cultural barriers to demonstrating how they can be navigated through concrete, community-engaged processes in resource-limited settings. Previous works [12] offer high-level frameworks but offer limited empirical evidence on the practical negotiation of trust, the role of local power structures like religious leadership, and the adaptation of AI tools to collective, rather than individual, decision-making models.

This paper addresses this gap by presenting an empirical case study of the AIMIX project in Kenya. We argue that the AIMIX project provides a concrete exemplar of “engaged implementation,” providing a real-world model for bridging the theory-practice divide. Through an analysis of rich qualitative data from 84 diverse stakeholders—including pregnant women, families, healthcare workers, and religious leaders—this study does not merely describe cultural challenges but demonstrates the specific strategies and adaptations required to build legitimacy for AI in a non-Western, high-context community. Ultimately, we contend that addressing the cultural gap in AI requires not just new frameworks, but a fundamental reorientation towards methodologies of co-creation, which this study details and defends.

2. Methods

This study employed a qualitative, multi-method approach, combining a narrative literature review with an in-depth, single-case study design [13] to explore the cultural and ethical dimensions of AI in healthcare, with a specific focus on the African context.

2.1. Literature Review Methodology

A narrative review was conducted to map the conceptual terrain of cultural and ethical factors influencing AI adoption in healthcare.

Search Strategy & Sources: Searches were performed in PubMed^® and Google Scholar between February-April 2025, covering publications from 2010-2025. This was supplemented by consultation with qualitative research experts and citation tracking. Foundational theoretical and methodological works published before 2010 (e.g., thematic analysis guidelines) were also consulted when relevant.

Search Terms: Combinations of the following keywords were used: “artificial intelligence,” “AI in healthcare,” “cultural competence,” “global health,” “ethics in AI,” “healthcare disparities,” “cultural determinants,” “local engagement,” “AI Africa,” “digital health.”

Inclusion Criteria: We included English-language peer-reviewed articles, grey literature, and policy reports that explicitly discussed cultural, ethical, or social aspects of AI in healthcare.

Exclusion Criteria: We excluded publications focusing solely on technical AI models without societal context, non-English publications, and non-analytical editorials.

Analytical Framework: From an initial pool of over 200 results, 68 sources were selected for final synthesis following the inclusion and exclusion criteria described above. Findings were organized and compared across four broad cultural perspectives—Western [14], Eastern [15], Global South [16], and Indigenous—to provide a structured global comparison before a deep dive into the African context.

2.2. A Review of Cultural Determinants of AI in Healthcare

To understand the challenges of implementing AI in diverse settings, it is essential to first map the global landscape of cultural determinants. This review moves beyond a simple catalogue of differences to analyze how deeply embedded cultural values—particularly the individualistic and technocratic assumptions of Western models shape the design, ethics, and anticipated adoption of AI tools. The following sections contrast these dominant Western frameworks with Eastern perspectives and, most critically, with the distinctive communal and relational paradigms of the African context, which fundamentally challenge the universality of the Western approach.

2.2.1. Global perspective

The global discourse on AI in healthcare has been predominantly shaped by Western and, to a significant extent, Eastern perspectives, each with its own internal diversity. A comparative analysis reveals how core cultural values create divergent expectations and ethical requirements for AI systems.

Cultural competence refers to the ability of researchers, practitioners, and institutions to understand, respect, and effectively engage with individuals and communities from diverse cultural backgrounds. [17] It involves recognizing the influence of cultural values, beliefs, practices, and social structures on human behaviour, interactions, and societal systems for conducting ethical, inclusive, and impactful research. Initiatives aimed at promoting inclusivity in AI design processes are becoming more prevalent, recognizing that culturally sensitive AI can lead to better health outcomes and patient satisfaction [18]. To comprehensively explore the impact of cultural differences on AI in healthcare across the world, we will provide a synthesised overview mostly based on social sciences criteria. It made possible to draw up a list of the main cultural differences to be taken into account depending on the context of design, development and adoption of AI tools in healthcare.

However, the conventional notion of cultural competency has been criticised for reducing culture to a static set of traits associated with ethnic or linguistic groups. This approach can lead to stereotyping and may overlook the individual patient’s context and lived experiences. We advocate for a more broad and dynamic understanding of culture, emphasizing, among others, the importance of the medical community engaging directly with patients to understand their unique perspectives and social determinants affecting their health.

2.2.2. African Perspective

The African context cannot be adequately understood as a mere variation of the Eastern or Global South perspectives; it presents a distinctive and powerful challenge to the Western-centric AI paradigm. The continent’s rich cultural diversity is underpinned by shared philosophical orientations and historical experiences that demand a fundamental rethinking of what constitutes “ethical” and “effective” AI. A focused examination reveals several themes central to AI adoption in African healthcare that directly contrast with the individualistic, technocratic assumptions of Western models.

Ubuntu and Communal Ethics in the Kenyan context [65]: Perhaps the most significant philosophical pillar comes from African humanist philosophies like Ubuntu, encapsulated in the principle “I am because we are.” Beyond a simple cultural motto, Ubuntu represents a relational conception of personhood in which individual identity is constituted through one’s obligations, reciprocity, and mutual care within the community. This philosophy finds direct expression in Kenya through concepts like Ujamaa (familyhood) and the national ethos of Harambee (all pull together). This stands in stark contrast to the Cartesian “I think, therefore I am” that underpins Western individualism [66]. Ubuntu emphasizes interconnectedness, communal well-being, and human dignity over individualistic gain [67]. Under this framework, an ethical intervention—including the deployment of an AI system—is judged by its capacity to foster social harmony, protect vulnerable members and strengthen the fabric of the community, rather than by technical performance alone. In practice, this means the success of an AI technology is judged not only by its diagnostic accuracy but by its ability to strengthen social cohesion, involve the community in its governance, and produce collective benefit [68]. This requires a shift from a model of “informed consent” focused on the individual to one of “communal endorsement”[69,70], where the legitimacy of a technology is negotiated with the community as a whole [71].

Infrastructural Realities and Frugal Innovation: The African context demands a move beyond technocratic solutions designed for stable, high-resource environments. Challenges like intermittent connectivity, limited electricity, and a scarcity of specialized healthcare workers are not mere obstacles to be overcome but are core design constraints [72,73]. This has spurred innovation in frugal AI, mobile-first (mHealth) solutions, and models that augment the capabilities of community health workers rather than replacing them, a direct contrast to the Western narrative of AI as a direct replacement for human expertise.

Data Sovereignty and Historical Legacies of Exploitation: Discussions about data in Africa are inextricably linked to histories of colonial and corporate extraction and exploitation. This has fuelled a powerful discourse on data sovereignty, demanding that data for AI be governed locally [74], stored on the continent where possible, and used in ways that directly and transparently benefit African populations [75]. Building trust requires data practices that actively counter this legacy of exploitation, going beyond the Western compliance-based model of GDPR to address deeper historical and political grievances.

The Centrality of Non-State Authority: As will be demonstrated in the AIMIX case study, formal regulatory frameworks are often only one part of the governance puzzle in many African contexts. The influence of religious leaders, tribal elders, and community health committees is profound and often surpasses that of distant government or corporate entities. Effective AI implementation requires engaging these non-state authorities as essential co-designers and moral gatekeepers. Their concerns about spiritual compatibility, modesty, and the dehumanization of care are not peripheral “cultural sensitivities” but central determinants of adoption, a reality that is largely absent from Western technocratic deployment models.

By centring these African perspectives, this review moves beyond applying global frameworks and instead builds a context-specific foundation that actively challenges the individualistic and technocratic assumptions of the Western model. Embedded in this rich, complex, and relational landscape, the AIMIX case study shows how these theoretical imperatives can be meaningfully operationalised to secure local legitimacy for AI. We therefore present a series of recommendations to guide such operationalisation (Table 1).

2.3. AIMIX Case Study Methodology

The Inclusive Artificial Intelligence for Accessible Medical Imaging Across Resource-Limited Settings (AIMIX) project in Kenya was selected as a critical case study exemplifying “engaged implementation” in a resource-limited African setting.

2.3.1. Case Study Context and Justification

The AIMIX project (Grant Agreement No. 101044779) focuses on developing affordable, AI-enabled ultrasound tools and building the social and ethical frameworks necessary for their successful adoption in prenatal care workflows in Kenya. It was selected for this research precisely because of this dual structure: AIMIX combines state-of-the-art AI model development—tailored to low-resource imaging conditions—with an equally strong emphasis on community engagement, cultural relevance, and responsible implementation. This integrated design provides a unique live setting in which to study not only how AI tools are technically developed for use in low-resource environments but also how socio-cultural, religious, and ethical factors are actively negotiated throughout their deployment process.

2.3.2. Data Collection

Multiple qualitative methods were used to collect rich, empirical data between August and December 2024.

Qualitative Interviews and Focus Group Discussions (FGDs): The team conducted a total of 58 qualitative sessions. This included 55 one-on-one, in-depth interviews and 3 Focus Group Discussions (FGDs). These activities engaged 84 unique participants. The sample comprised a diverse range of stakeholders: pregnant women, their partners, healthcare providers, community health promoters, Sub-County Health Management Teams, and community opinion leaders (including local administrators, religious leaders, women, and youth leaders, and a traditional birth attendant). The interviews and FGDs explored socio-cultural contexts, ethical considerations, and perceived needs and barriers related to AI-assisted prenatal ultrasound.

Community Meetings: The project organized 47 community meetings in Rabai, engaging pregnant women, their husbands, and community health volunteers. These meetings served as an open forum to discuss expectations, concerns, and feedback on the AI-based healthcare tools.

Stakeholder Workshops: A series of workshops were held with local authorities, including Sub-County Health Management Teams, local administration (chiefs, village elders), and religious leaders. These workshops, held between September and November 2024, focused on integrating the AI tools within existing maternal health programs and local governance structures.

Community Advisory Board (CAB): A CAB was established in September 2024, comprising 17 members from the local Rabai community (including mothers, husbands, and healthcare workers). The CAB met regularly to provide ongoing feedback, ensure community alignment, and strengthen researcher-community trust.

Document Analysis: The study also analysed internal AIMIX project documents, technical reports, and community engagement summaries produced between August and December 2024 to provide complementary context on project design and adaptation processes.

2.3.3. Data Analysis

All qualitative data (interview and FGD transcripts, workshop notes, meeting minutes) were analysed using a thematic analysis approach [76]. The process involved familiarization with the data, generating initial codes, searching for themes, reviewing themes, and defining and naming themes. This process allowed for the identification of recurring patterns related to cultural beliefs, gender norms, trust in technology, and community perceptions of AI.

2.3.4. Ethical Considerations

The AIMIX project obtained full ethical approval from the Aga Khan University Institutional Scientific and Ethics Review Committee (ISERC), followed by research permits from the National Commission for Science, Technology, and Innovation (NACOSTI), and local authorization from the Kilifi County Department of Health. Ethical approval was also granted by the University of Barcelona Ethics Committee. The study fully complied with established human subjects research principles, ensuring informed consent, strict confidentiality, and robust data protection measures.

3. Results: Navigating the Social Ecosystem for AI Legitimacy

The analysis of the AIMIX case study reveals that the acceptance of AI was not determined by technical performance alone but emerged from a complex, socially negotiated process. Communities evaluated the technology through existing moral, religious, familial, and governance structures that shape how prenatal care is understood and practiced. This directly challenges the dominant “deployment” narrative in global health AI, which often assumes that once a tool is validated, adoption will naturally follow. Instead, AIMIX demonstrated that legitimacy must be earned across multiple socio-cultural dimensions—ranging from religious norms and gender dynamics to health literacy, trust in institutions, and regulatory expectations. Table 2 summarises the twelve key socio-cultural factors identified through the case study and outlines how AIMIX operationalised culturally competent strategies to address each of them.

3.1. The Moral Gatekeepers: Religious Leaders as Co-designers

Our engagement revealed that religious leaders acted not as passive stakeholders, but as essential moral and ethical co-designers, a finding that challenges the common treatment of “culture” as a peripheral variable in ethical AI frameworks [6,11]. Their primary concern transcended common issues like data privacy, focusing instead on a profound theological unease: the potential for AI to disrupt the sacred, compassionate covenant of the doctor-patient relationship.

From Suspicion to Sanction: The project’s initial, technocratic framing of AI as an “efficiency tool” was met with deep scepticism. This forced a fundamental re-negotiation of the technology’s moral purpose. We subsequently articulated a theological rationale for assistive AI, positioning it not as a replacement, but as a tool that “frees the healthcare provider to offer more compassionate, human care.” This reframing was crucial for gaining their endorsement.

Proactive Ethical Safeguarding: The leaders’ insistence on withholding fetal sex information was not a simple preference but a community-level ethical safeguard. They foresaw a social risk—gender-based selective abortion—that the technical team had not prioritized. This intervention demonstrates how localized ethical foresight can surpass the generic principles of “non-maleficence” found in global guidelines [3], actively shaping the AI system to preempt societal harm.

The “Sermon as Sensitization” Model: The subsequent endorsement from religious leaders, delivered through sermons and community talks, carried a legitimacy no project brochure could match. This illustrates a key finding: in this context, trust is transferred, not built from scratch. The AI gained credibility by being sanctioned by a trusted moral institution, a mechanism of legitimation largely absent from Western models of technology adoption.

3.2. The Collective End-User: Re-inventing Consent within the Family Unit

The data starkly contrasted the Western model of individual patient autonomy, which underpins most bioethical frameworks and “informed consent” protocols [52]. We found that for technologies impacting pregnancy, the family unit is the primary decision-making entity.

The “Husband’s Veto” and Familial Authorization: Concerns from husbands about data privacy were, in practice, a manifestation of their role as family protectors and decision-makers. Their buy-in was a non-negotiable social prerequisite for their wives’ participation. This necessitated a move from a model of “informed consent” to one of “familial authorization.” This finding directly challenges the universality of individual autonomy and suggests that in strong-structure contexts, ethical AI requires a relational, rather than individualistic, model of consent.

Redefining “The User”: The common Human-Centered Design (HCD) focus on the “pregnant woman” was revealed to be insufficient. Our findings show that in this collective structure, the true “user” is a network—the woman, her spouse, and often her mother-in-law. AI tools designed for individual self-management were destined to fail. This necessitates a fundamental shift from a Western human-computer interaction paradigm to a community-computer interaction paradigm, requiring features that support and respect collective decision-making dynamics.

3.3. The Governance Bridge: Local Authorities and Systemic Integration

Engagement with local authorities revealed that for AI to be sustainable, it must be woven into the fabric of existing governance, countering the tendency for top-down digital health projects to create parallel systems [72].

From “Project” to “Programme”: Health management officials were less interested in the AI’s algorithm and more in its integration pathway into their existing maternal health programs. Their primary question was: “How does this become our tool, not your experiment?” This perspective led to the co-development of training materials and reporting pathways that aligned directly with county health priorities, ensuring the project enhanced, rather than bypassed, local systems.

The Tripartite Model of Legitimacy: The most effective engagement structure that emerged was a tripartite model, linking the project simultaneously to community representatives (CAB), religious authorities, and the local health administration. This created a robust system of checks and balances where social, moral, and bureaucratic legitimacy reinforced one another. This model offers a concrete alternative to the often abstract “multi-stakeholder” engagement promoted in global health literature, providing a specific architecture for accountability in non-Western settings.

4. Discussion: From Western Frameworks to Socially-Accountable AI

The AIMIX case study provides a detailed blueprint for what “ethical AI” means in contexts where community and authority are distributed and relational. Our findings directly challenge the portability of Western-centric models and demand a more nuanced, socially-accountable paradigm.

4.1. The Limits of “Trustworthy AI” in a Relational World

Prevailing frameworks for “Trustworthy AI”[5,11] prioritize transparency, fairness, and accountability to the individual user. The AIMIX experience exposes the limitations of this model when applied to communal societies.

Our finding that informed consent must be re-engineered as “familial authorization” directly challenges the Western bioethical principle of individual autonomy as a universal standard. In our context, the Western concept of autonomy clashed with the local reality of familial interdependence. This novel concept is a critical addition to the ethical AI lexicon, suggesting that consent models must be fluid and adaptable to the social structures they encounter.

We found that abstract ethical principles are insufficient and require cultural translation. “Fairness” or “non-maleficence” must be translated into locally resonant moral terms. For religious leaders, this meant framing AI through the lens of compassion and human dignity; for families, it was about collective well-being and protection. This finding challenges the checklist approach of many global AI ethics guidelines [5] and instead suggests that AI ethics must be a dialogical process of cultural translation, where universal principles are given life through local values.

4.2. The Specificity of Engagement in Strong-Structure Contexts

A key question concerns what is distinctive about engaging communities shaped by strong family and religious structures. This study provides a clear, three-part answer that contrasts with common Western assumptions:

The “Why” is rooted in social hierarchy: the “why” behind the concerns—the fear of dehumanization, the need for familial authorization—is rooted in the primacy of relational harmony and predefined social roles. This stands in stark contrast to the individualistic risk-benefit calculus often assumed in Western HCD. Authority is not evenly distributed; it is vested in specific figures (husbands, elders, clergy). Effective engagement requires mapping and respectfully engaging this hierarchy, not attempting to circumvent it.

Effective implementation required the team to cede a degree of control: the most significant adaptations—withholding fetal sex, creating parallel engagement sessions for men—were not our ideas. They were concessions to community logic. This finding challenges the technocratic assumption that developers retain control over system design and implementation. It demonstrates that true co-creation in these contexts involves ceding a degree of technical and procedural control to social and ethical imperatives.

Legitimacy is borrowed, not built: the project’s legitimacy was not built solely on its technical merits but was “borrowed” from the pre-existing legitimacy of religious and local authorities. This turns the standard tech deployment model on its head. Whereas in Western contexts, trust might be built through technical transparency and corporate reputation, our study shows that in contexts like rural Kenya, the most important factor for adoption may be the quality of the technology’s endorsement by the community’s moral and governance pillars.

4.3. Contribution: Operationalizing “Local Legitimacy” as an Alternative Framework

The primary contribution of this research is to move the discourse from acknowledging the importance of culture to operationalizing it. We provide empirical evidence that “local legitimacy” is a tangible, achievable outcome built on three pillars that collectively form an alternative to Western-centric models:

Moral Legitimacy: secured not through ethics review boards alone, but through deep engagement with religious leaders, translating AI ethics into local theological frameworks. This challenges the secular, principle-based foundation of most AI ethics.

Social Legitimacy: achieved by designing for the family unit, not the individual, and re-engineering consent as a collective process. This challenges the individual-as-end-user paradigm at the heart of both bioethics and human-centered design.

Bureaucratic Legitimacy: ensured by embedding the technology within existing local government and health management systems. This challenges the “plug-and-play” or “disruptive” innovation model often favoured by technology firms.

This tripartite model offers a replicable framework for researchers and policymakers. It argues that the budget, timeline, and success metrics for AI deployment in the Global South must account for the intensive work of building this multi-faceted legitimacy. The AIMIX case proves that a technically sound AI tool is a scientific achievement; a culturally legitimate AI tool is a social one, and in many parts of the world, the latter is the prerequisite for the former to matter.

5. Conclusions

This study demonstrates that for AI to become a truly transformative force in global health, a fundamental shift from top-down implementation to co-created legitimacy is required. The in-depth case study of the AIMIX project in Kenya provides a granular, empirical model for this shift, moving beyond the identification of cultural barriers to reveal the precise mechanisms for overcoming them.

Our findings yield a critical, field-tested insight: in contexts where authority is distributed and relational, the “end-user” of an AI tool is not an individual patient but the entire social ecosystem—a finding that fundamentally challenges the individualistic orientation of Western bioethics and HCD. The AIMIX data uniquely show that trust was built not through explaining the algorithm’s accuracy, but by navigating a network of moral and social gatekeepers. This involved:

• Translating “Ethical AI” into Local Moral Frameworks, thereby challenging the sufficiency of secular, principle-based ethics guidelines.
• Re-engineering “Informed Consent” as “Familial Authorization,” thereby offering a concrete alternative to the universal application of individual autonomy.
• Embedding AI within Existing Community Governance, thereby proposing a sustainability model that counters top-down, disruptive deployment strategies.

Therefore, this research does not merely add a “cultural” caveat to existing AI deployment models. It provides an empirically-grounded challenge to their foundational assumptions and offers a replicable pathway for designing AI systems that are accountable to the communities they serve, ensuring that the pursuit of health equity drives technological innovation, not the other way around.