Abstract
Background
The integration of Artificial Intelligence (AI) in healthcare promises significant advancements in patient care, yet its success heavily relies on public trust and acceptance, particularly in diverse socio-economic contexts like South Africa. This study investigates South African residents’ willingness to trust AI in healthcare decisions, exploring the impact of socio-demographic factors such as age and religion on their preferences.
Methods
Utilizing a cross-sectional online survey distributed via Facebook, we gathered data from 341 respondents across South Africa. The survey assessed participants’ preference for human versus AI doctors in serious health scenarios, alongside demographic information including age, gender, educational attainment, religion, and home language. Weighting adjustments were applied to align the sample with South Africa’s demographic proportions for home language and gender. Data were analysed to explore correlations between these demographics and preferences for AI in healthcare.
Results
A significant majority (73.7% weighted) expressed a preference for a human doctor over an AI doctor. Notably, the importance of religion (p < .001) and specific age groups (p = .025) significantly influenced preferences. A significant proportion of respondents for whom religion was “not too important,” as well as those in the 40–49 age group, preferred an AI doctor.
Conclusions
This study underscores the need for innovative governance models tailored to resource-constrained settings, where traditional human-in-the-loop requirements may not always be feasible. Future research should explore how socio-cultural factors and trust dynamics influence public attitudes toward AI in healthcare and investigate models that ensure safety and accountability while addressing practical limitations in healthcare delivery systems.
Supplementary Information
The online version contains supplementary material available at 10.1186/s12910-025-01272-8.
Keywords: Artificial intelligence, Healthcare, Public trust, South Africa, Socio-demographic factors
Introduction
The advent of artificial intelligence (AI) in healthcare represents one of the most significant technological milestones of the 21st century, promising to revolutionise patient care, diagnostics, treatment planning, and patient monitoring [1–3]. In various regions worldwide, including sub-Saharan Africa, the integration of artificial intelligence into healthcare settings is advancing at varying paces, driven by its potential to improve healthcare outcomes and enhance care delivery efficiency, though barriers to widespread adoption and consistent cost reductions remain significant [4–8]. However, the successful adoption and effective utilisation of AI in healthcare depend on the trust and acceptance of these systems by healthcare professionals and the public, underpinned by robust governance frameworks [9–14].
In South Africa, a country marked by stark contrasts in healthcare access, quality, and resource distribution, the implications of AI adoption in healthcare are profound. South Africa’s healthcare system is bifurcated into a well-resourced private sector and an under-resourced public sector [15, 16]. This dichotomy presents unique challenges and opportunities for the implementation of AI technologies. On one hand, AI has the potential to mitigate some of the resource constraints in the public sector by improving diagnostic accuracy, enabling remote patient monitoring, and facilitating patient data management. On the other hand, the disparities in access to technology, digital literacy, and healthcare infrastructure raise critical questions about the equitable deployment and utilisation of AI across different segments of the South African population.
This study sought to investigate the willingness of South Africans to trust AI in making healthcare decisions in the context of serious illness. Specifically, it addresses the following research questions: (1) To what extent do South Africans trust AI in high-stakes healthcare decision-making? (2) What socio-demographic factors, such as age, gender, education, religion, and home language, influence preferences for AI versus human doctors in these scenarios? By focusing on serious illness, the study prioritises contexts where trust and perceived competence are paramount, ensuring that responses reflect participants’ attitudes toward critical healthcare scenarios rather than lower-stakes situations. As an initial exploration, this inquiry aims to provide a starting point for understanding societal readiness for AI in healthcare in South Africa. By situating the research in the South African context, the study contributes insights into how diverse populations perceive and accept the role of AI in life-critical healthcare decisions, offering a foundation for more detailed investigations into policy, system design, and education strategies for AI integration.
Literature review
This literature review details main themes found in the literature on patient opinions on artificial intelligence (AI) use in healthcare. We begin by considering patient attitudes to AI use in healthcare generally, mentioning some demographic determinants of AI trust. Next, we consider sources of trust in healthcare and AI specific challenges to that trust. Lastly, we describe patient opinions on AI’s impact on the doctor-patient relationship.
Patient attitudes
A number of studies from the United States and Europe report mixed attitudes towards AI use in healthcare. Some studies indicate positive attitudes from patients, particularly regarding the potential for increased accuracy and efficiency in clinical decision-making, although these participants also indicate some wariness towards AI use. [17, 18]. In particular, oncology patients in the United States and the United Kingdom are hopeful that AI can improve the speed and quality of data analysis and help detect mistakes, especially in the context of cancer care [19]. Additionally, in Europe, 63.4% of respondents report approving or strongly approving of AI use [20].
Americans also generally perceive healthcare as a field where AI applications could bring benefits, particularly in advancing public health, but they express caution about AI making personal health decisions and managing healthcare data [21]. Although some American participants expressed optimism about AI improving healthcare, this enthusiasm was moderated by significant discomfort and doubt regarding potential negative impacts, including privacy risks, reduced clinician interaction, increased healthcare costs, and AI’s lack of explainability [22].
Despite some positive perceptions, studies consistently highlight significant apprehensions and discomfort among patients, especially when AI systems lack transparency or explainability [22–24]. In particular, there is considerable discomfort among American patients when AI is relied on by health practitioners for their medical care, despite some potential benefits like reducing bias in healthcare [24].
When it comes to personal medical decisions, caution dominates. Two studies from the United States found that participants were generally hesitant to rely on AI systems in these contexts. Specifically, Tyson et al. [24] found that 60% of Americans would feel uncomfortable with their healthcare provider relying on AI for diagnosis and treatment, while Beets et al. [21] reported that only 8% of participants were comfortable with an AI making end-of-life care decisions.
While positive attitudes towards AI do exist, they do not necessarily lead to higher AI adoption rates. A Canadian study found that positive attitudes did not correlate with increased AI uptake, nor did negative attitudes invariably result in resistance [25]. A similar trend was observed in the use of surgical robots, where American participants generally viewed the technology positively but became more cautious when considering its use for their own treatment [24]. More determinative of AI adoption are factors such as familiarity with the technology and education. Studies from the United States and the United Kingdom show that participants unfamiliar with AI-powered surgical robots were more apprehensive about using them for personal care, whereas those familiar with AI were either evenly divided or supportive of their use [19, 24].
Education emerged as a key determinant of support for AI use in healthcare in studies conducted in the United States, United Kingdom, and Europe. Higher educational attainment and income levels were positively associated with support for AI applications in healthcare [19, 24]. Specific education on AI in healthcare increased trust among European participants [20]. Additionally, digital literacy and familiarity with AI technology were positively associated with acceptance in healthcare settings [26]. Education on AI in healthcare also led to increased willingness to share data for AI-driven health research among American participants [21].
Comparison of trust in AI vs. trust in humans
Institutions and professionals are an important source of trust for AI in healthcare. A study from the United States found that participants were more likely to place trust in AI systems when physicians endorsed or recommended their use [27]. Similarly, a Canadian study found that AI applications aligning with appropriate regulation and approved by regulatory bodies were more acceptable to participants [28].
In studies in the United States and Europe, the least trusted stakeholders were commercial entities [20, 29]. Participants showed concern over the safety of their data where private stakeholders have access in Canadian and Australian studies [25, 30]. Thisconcern may explain why some participants in a further study from the United States to favour physical consultations with their physicians, showing the trust they have in their physicians [31].
An important means of building trust was knowing how an AI came to its decision. One study from the United States suggested that a lack of interpretability may produce lower trust in AI recommendations [31]. However, when American and Australian participants were asked to prioritise aspects of AI in healthcare, accuracy was valued much higher than interpretability [30, 32], being one of the most important factors determining AI use [33]. American participants considered AI accuracy crucial in its ability to access more data than humans [33]. Similarly, in another study with American participants, they were substantially more likely to select AI systems where they are proven to be more accurate than humans [28].
Importantly, the prioritisation of accuracy over interpretability became more pronounced as the stakes of AI decisions were higher and resources were scarce [32]. But comfort of AI use varied depending on the clinical application of the system [22, 24]. Accordingly, studies from the United States and the United Kingdom found AI system acceptance was lower as severity of the disease being assessed increased and resources became more scarce [22, 26]. Therefore, American patients were comfortable with AI recommendations in low-risk interventions such as general wellness strategies or talk therapy; however, they were uncomfortable with AI systems diagnosing disease or recommending medication [18]. This may be a consequence of mixed opinions of participants on AI impact on healthcare outcomes. Whilst some studies in the United States reported positive attitudes to AI impact [22], others reported ethical concerns and a significant lack of support for AI use improving healthcare outcomes [21, 22, 34].
The mixed perceptions of AI impact on healthcare outcomes are further illustrated in a study from the United Kingdom in participants’ concern for AI becoming a source of error [19]. This concern was founded in American participants in the perceived rapid emergence and deployment of AI technologies in healthcare and the worry that current issues will be exacerbated without fully understanding how to remedy them. A study in the United States found that this aversion proved so persistent that, as a result, AI uptake may not increase even where AI is proven to be accurate and a physician is given the final decision on medical care [27].
AI impact on the patient-physician relationship
A key healthcare concern which American participants raised was the disruption which AI systems may make to the patient-physician relationship [24]. The patient-physician relationship is a primary source of trust in healthcare, and participants expressed discomfort with the possibility of AI interfering with this bond, especially in mental healthcare settings [18]. In a study from the United Kingdom, radiotherapy patients, particularly, were concerned with potentially less personal interaction and fewer opportunities to raise concerns or have worries assuaged by another human [35]. AI’s inability to embody crucial criteria of human relationship, empathy, and warmth led European participants to believe that AI could not replace humans in healthcare [17]. Aligning with this, European patients generally trusted physicians over AI in most clinical settings except when considering the most current clinical knowledge and generally trusted AI most when it was under the supervision of a physician [36].
Relegating the AI system to an advisory role was not a complete solution though. Where AI merely provided recommendations, participants in a study from Europe showed concern at these recommendations not being challenged, leading to overreliance and loss of expertise or blind trust in AI systems [17].
In the interest of upkeeping responsibility for decision, American, European, and Australian participants stated an interest in preserving their choice to use AI systems and know when AI systems are being used [17, 30, 33]. This maintains a means for patients to dispute AI decisions or correct recommendations [17]. In a study from Europe, not all participants considered disclosure appropriate though, some suggesting that it may be unnecessary, confusing and overwhelming and further erode trust or cause harm [17]. Further, in a study from Australia although some participants were empowered by the ability to challenge results, others argued that they did not challenge current results from their physicians, and therefore did not consider it important to be able to challenge AI recommendations [30]. Nevertheless, the majority of participants recognised knowing who is responsible for decision-making in healthcare as foundational to patient challenging and accepting of decisions [30].
Importantly, in a study from the United States where an AI system and a physician conflicted in their decision, participants reported being more likely to trust the physician’s decision, even where AI systems and physicians were equally effective [33]. This was echoed in determinations of who should generally make final decisions, with most participants in studies from the United States, Canada and Europe agreeing that physicians should make all care decisions [17, 37]. At least one study from the United States suggested this question may not be significant as they found that whether the physicians merely deferred to AI recommendations or incorporated them into their care did not affect acceptance of AI recommendations [27].
Ultimately, American patients strongly preferred supervised AI use, considering AI as a means to doublecheck or compliment physicians’ efforts instead of a stand-alone technology [33]. They were unwilling to undergo procedures such as autonomous robotic surgery without immediate human supervision [38]. Even where they were willing, it was only where the surgeon had fully explained the exact application of the AI system in the surgery [33].
Ethical considerations in AI adoption in healthcare settings in Africa
The adoption of artificial intelligence (AI) in healthcare across Africa involves complex ethical considerations that intersect with cultural, religious, and socio-economic factors. Elendu et al. [39] highlight how societal trust in AI extends beyond technical reliability to encompass alignment with moral and cultural values. In a similar vein, Ferlito et al. [40] suggest that community-based ethics, such as Ubuntu—which emphasise interconnectedness and shared responsibility—offer an important foundation for ethical AI governance in Africa. Public concerns around data misuse, algorithmic bias, and the dehumanisation of healthcare decisions are central to these considerations, necessitating context-sensitive approaches.
Naidoo et al. [41] focus on the gaps in South Africa’s existing legal frameworks for AI adoption, identifying biases in data collection and outdated regulations as significant barriers to public trust. They advocate for modernising policy to address these gaps, proposing a national governance framework that incorporates fairness and accountability while empowering healthcare workers through reskilling initiatives. These steps are essential to bridging the divide between technological advancement and societal acceptance.
Sihlahla et al. [42] explore ethical principles governing AI in South African radiology, emphasising the need to mitigate algorithmic biases and ensure equitable access to AI benefits. Their work underscores the importance of culturally responsive governance to build trust in AI applications. In a broader analysis, Townsend et al. [43] examine AI governance in 12 African countries, highlighting the lack of public engagement and ethical oversight as barriers to effective implementation. Their findings reveal the need for inclusive policymaking that incorporates diverse cultural and ethical perspectives, fostering trust and acceptance among communities.
Eke et al. [44] echo these sentiments by stressing the importance of public engagement in shaping AI policies. They propose integrating African philosophical values, such as Ubuntu, into governance models to ensure ethical alignment and societal trust. These recommendations align with global frameworks like those of the World Health Organization, which advocate for transparency, accountability, and cultural sensitivity in AI deployment.
Kabata and Thaldar [45] highlight the challenges of implementing the human-in-the-loop requirement in low-resource settings, such as rural Africa, where medical expertise is often limited. They propose a human-rights-based regulatory framework prioritising accessibility and safety, shifting oversight to public institutions to ensure accountable AI governance. Their emphasis on ethical principles, including autonomy and beneficence, underscores the importance of culturally sensitive approaches in building public trust.
In conclusion, the effective adoption of AI in African healthcare requires governance frameworks that integrate cultural values, address biases, and incorporate public engagement. A human-rights-based approach, as advocated by Kabata and Thaldar [45], provides a promising pathway for fostering trust and ensuring that AI technologies align with the needs of diverse and resource-constrained settings.
Methodology
Ethics approval
Ethical approval was obtained from the Biomedical Research Ethics Committee of the University of KwaZulu-Natal (protocol reference number BREC/00006176/2023). Informed consent was provided electronically, as detailed in our discussion below. In general, the study adhered to the Declaration of Helsinki.
Sampling method
The study recruited respondents via Facebook, displayed to people indicating that they currently reside in South Africa. As of January 2024, 75% of South Africans were internet users, with Facebook being used at least once a month by 88% of internet users aged 16 to 65—second only to WhatsApp at 94% [46]. The next closest competitor, TikTok, had a usage rate of 74% [46]. However, since advertisements cannot be purchased on WhatsApp, Facebook emerges as the social media platform with the largest advertisement reach in South Africa, making it an ideal choice for recruitment.
A recruitment video of about 6 s appeared in the mobile newsfeed of Facebook users who were 18 or over and resided in South Africa. The video storyboard is presented in supplementary file S1. It ends with the question: “Do you want to take part in the forum?” Upon clicking “Yes”, the person was directed to the Google Forms URL where the survey was hosted.
On the survey landing page, prospective respondents were provided with information about the survey, and requested to consent to participate in the survey. No renumeration was offered. By clicking “Next” to start the survey, respondents indicated and electronically recorded their consent to participate in the study.
Potential limitations of the sampling method
The use of Facebook for recruitment offers several advantages, such as the ability to reach a large and geographically diverse audience. However, it also introduces potential limitations that must be acknowledged.
Internet access disparities: The reliance on an online platform inherently excludes individuals without internet access.
Self-selection bias: Participation in the survey was voluntary, which might have resulted in self-selection bias. Individuals who chose to respond may have had stronger opinions or greater interest in the topic of AI in healthcare, potentially limiting the diversity of perspectives captured in the survey.
Cultural representativeness: While Facebook allows targeting by geographic location, it does not account for the cultural, linguistic, and religious diversity of South Africa. This limitation underscores the importance of weighting adjustments to better align the sample demographics with national statistics, which we discuss below.
Measures
The survey was developed to explore preferences for human versus AI doctors in the context of serious illness, along with the influence of socio-demographic factors. The questions were carefully designed to align with the study’s objectives and were informed by existing literature on AI adoption in healthcare. The six survey questions are presented in Panel I. All six the questions were multiple choice.
| Panel I: The survey questions |
| If you are seriously ill, who would you rather trust to diagnose your illness and prescribe the appropriate treatment? |
| ∙ A human doctor |
| ∙ An AI doctor |
| What is your gender? |
| ∙ Male |
| ∙ Female |
| ∙ Prefer not to state |
| How old are you? |
| ∙ 18 or 19 years old |
| ∙ 20s |
| ∙ 30s |
| ∙ 40s |
| ∙ 50s |
| ∙ 60s or older |
| What is your highest educational qualification? |
| ∙ Grade 7 or no education |
| ∙ Grade 10 |
| ∙ Grade 12 |
| ∙ College diploma |
| ∙ University degree |
| How important is religion in your life? |
| ∙ Very important |
| ∙ Somewhat important |
| ∙ Not too important |
| ∙ Not at all important |
| What is your home language? |
| ∙ Afrikaans |
| ∙ English |
| ∙ Ndebele |
| ∙ Pedi |
| ∙ Sotho |
| ∙ Swati |
| ∙ Tsonga |
| ∙ Tswana |
| ∙ Venda |
| ∙ Xhosa |
| ∙ Zulu |
| ∙ Other |
Development of the AI question
The key question asking participants to choose between a human doctor and an AI doctor was crafted to assess trust in AI in high-stakes healthcare scenarios. The question intentionally presented a binary choice to prompt respondents to make a clear decision, avoiding the introduction of complexities such as human-AI collaborations, which could imply a hierarchical relationship and distract from the core objective. By leaving the term “AI doctor” undefined, the question allowed respondents to rely on their intuitive understanding of AI technology, reflecting real-world scenarios where patients may not receive detailed explanations of the tools being used.
Survey language
The survey was administered in English only, a decision informed by practical constraints and the expectation that the target population—Facebook users in South Africa—would have a sufficient level of English proficiency. Future iterations could benefit from offering the survey in multiple South African languages to improve inclusivity.
Socio-demographic variables
The socio-demographic questions (gender, age, educational attainment, religion, and home language) were selected based on prior studies that identified these variables as potentially influencing attitudes toward AI in healthcare. Including these variables enabled the study to assess trends and patterns specific to the South African context.
Gender
Gender influences attitudes toward AI use in healthcare, though findings vary. Parry et al., [38] in a study across urban and rural United States health centres, found that male participants were significantly more comfortable with AI in orthopaedic care than females (p =.0001). Similarly, Tyson et al., [24] in a United States-wide survey, reported that 47% of men were comfortable with AI-based robots for their own surgery, compared to 33% of women. York et al., [47] conducted at a London, United Kingdom, teaching hospital, also found that male participants were significantly more confident in AI-assisted skeletal radiograph interpretation than their female counterparts (p<0.001), though specific proportions were not reported.
Age
Literature on age as a demographic determinant of AI uptake in healthcare reveals contradictory findings. Parry et al. [38] (in the United States) found that average AI comfort levels in orthopaedic care differed significantly by age (p =.032), with participants aged 55 to 75 years being most comfortable (mean: 6.8/10) and those aged 18 to 55 years least comfortable (mean: 6.0/10). This was contradicted by York et al. [47] (in the United Kingdom) found younger patients had slightly greater confidence in AI-assisted skeletal radiograph interpretation (r=-.170, p =.0123). Similarly, Robertson et al., [27] in a United States-wide survey, showed older respondents were less likely to choose AI over human physicians (OR: 0.99, p =.03). Tyson et al., [24] also United States-wide, found younger adults were more open to AI in healthcare, particularly for applications like skin cancer screening.
Educational attainment
Educational attainment significantly influences attitudes toward AI adoption in healthcare. Participants with higher levels of education are more comfortable with AI technology in their care, according to Tyson et al. [24] (in the United States), and are more likely to approve of AI systems as part of their orthopaedic care team, as reported by Parry et al. [38] (also in the United States). Tyson et al. further note that those with advanced education, such as postgraduate degrees, are more optimistic about AI’s potential to improve patient outcomes. York et al. [47] (in the United Kingdom) support these findings to an extent, observing that higher education correlates with greater confidence in AI-assisted skeletal radiograph interpretation, though their study is limited to this specific application.
Religious views
Religious beliefs can influence patients’ acceptance of AI in healthcare, though research on this topic remains limited and context-specific. Robertson et al. [27] (in the United States) found that respondents who viewed religion as important were significantly less likely to choose an AI clinic over a human physician for diagnosis (OR = 0.64, CI: 0.52–0.77, p <.001). Their qualitative interviews suggested some patients prefer human physicians, believing God works through them. In African contexts, Obasa [48] suggests religious beliefs, as part of cultural worldviews, may shape AI perceptions by emphasising communal values and trust in human caregivers. For example, moral understanding, potentially informed by religion, can guide ethical AI development through phronesis. However, Obasa’s work lacks empirical data on religion’s direct impact. The scarcity of studies analysing religious influences, particularly outside Western settings, underscores the need for further research to understand how diverse religious beliefs affect AI adoption in healthcare globally.
Cultural markers
The literature largely details cultural differences in relation to AI adoption by measuring the races of participants. Studies conducted in the United States consistently demonstrate racial variations in attitudes toward AI in healthcare. Tyson et al. [24] found that although overall Black adults were more optimistic about AI reducing racial/ethnic bias in healthcare, they were less optimistic compared to their White and Hispanic counterparts (40% versus 54% for White and 50% for Hispanic adults), suggesting lower receptivity to AI applications such as skin cancer screening. Similarly, Robertson et al. [27] found that Black respondents had lower odds of selecting an AI clinic for diagnosis (OR = 0.73, p =.023), while Native American respondents had higher odds (OR = 1.37, p =.041) compared to White respondents. Supporting this trend, Khullar et al. [22] reported that non-White respondents expressed greater concern about AI-related misdiagnosis (46.9% versus 35.9% for Whites), privacy (38.6% versus 27.2%), and costs (45.4% versus 30.3%), indicating higher discomfort with AI use in healthcare compared to White respondents.
The association between race and culture may be viewed differently in various countries. In South Africa, we opted to use home language as a cultural marker rather than race. The term “Bantu language” refers to a group of languages within the Niger-Congo family, widely spoken across sub-Saharan Africa. In South Africa, Bantu languages include Zulu, Xhosa, Sotho, and others, which are predominantly spoken by Black South Africans. Using “Bantu language” as a cultural marker reflects linguistic diversity while avoiding racial categorisation, aligning with the country’s unique sociolinguistic and historical context.
Data analysis
The primary objective of the data analysis was to determine participants’ preferences for human versus AI doctors in the context of serious illness and to explore how socio-demographic factors influenced these preferences. The analysis involved preparing the data for interpretation through cleaning, weighting, and statistical evaluation.
Weighting adjustments
To ensure the survey results better reflected South Africa’s population demographics, weighting adjustments were applied. These adjustments accounted for two key demographic variables:
Home language: The weights were derived based on the proportions reported in the most recent South African census, particularly aligning the proportion of South African Bantu home languages (78.5%) [49] to represent the national population more accurately.
Gender: Gender weights were applied to represent the national distribution between men and women (48.5% versus 51.5%) [49], addressing the overrepresentation of female respondents in the raw data. Because the weighting was based on official South African census data, which includes only the categories male and female, respondents needed to select one of these two options to be included in the weighted analysis. Accordingly, stating a gender became an inclusion criterion.
Weights were calculated using post-stratification techniques. Each respondent’s raw response was multiplied by a weight corresponding to the inverse of their over- or underrepresentation in the raw sample. These adjusted weights were used in all subsequent analyses to enhance the generalisability and validity of the findings.
Descriptive statistics
Descriptive statistics were calculated to summarise the sample’s demographic characteristics and response distributions. Frequencies and percentages were used to provide an overview of trends in preferences for human versus AI doctors and to describe the composition of the weighted sample in terms of age, gender, home language, and education.
Inferential statistics
To investigate relationships between preferences for human versus AI doctors and socio-demographic factors, the following statistical tests were conducted:
Chi-Square Tests: These tests assessed associations between categorical variables such as gender, age, education, and home language, and participants’ preferences.
Logistic Regression Analysis: A logistic regression model was employed to determine the relative influence of socio-demographic factors on the likelihood of preferring an AI doctor over a human doctor.
All statistical tests were conducted at a significance level of p <.05.
Software used
All data analysis was performed using SPSS, which facilitated the calculation of weights, statistical testing, and visualisation of results.
Representativeness and limitations
Although weighting adjustments improved the representativeness of the sample, the recruitment process’s reliance on Facebook ads may have introduced biases related to internet and social media access. These limitations were considered when interpreting the results, acknowledging that the findings serve as an initial exploration of societal readiness for AI in healthcare.
Results
The responses are contained in supplementary file S2. 341 respondents participated in the survey between October 2023 and March 2024. Participants were recruited through targeted advertisements on Facebook. While our initial goal was to achieve a larger sample size, the recruitment process progressed slower than anticipated. To maintain momentum, additional advertisements were purchased. Due to budgetary constraints, recruitment had to conclude when the sample size reached 341 participants.
After weighting adjustments were applied to address the underrepresentation of male respondents and speakers of South African Bantu home languages, the final weighted dataset comprised 336 respondents. The results that follow are based on the analysis of this weighted dataset.
Although the sample size was influenced by practical limitations, it exceeds the required sample size of 255 as calculated with a margin of error of 0.05, an alpha value of 0.05 and a p value of 0.79 as found in this study.
The demographic breakdown of the 336 respondents is presented in Table 1. Some highlights are as follows: The overwhelming majority were women; the respondents’ ages were distributed over all the defined age groups; most respondents finished secondary school (grade 12), and a third of the respondents had some form of tertiary education; regarding the importance of religion, two-thirds of the respondents indicated that it is very important to them; and with respect to home language, only about a quarter of the respondents indicated a South African Bantu home language.
Table 1.
Socio-demographic profile
| Variable | Categories | n | % |
|---|---|---|---|
| Gender | Male | 66 | 19.6 |
| Female | 270 | 80.4 | |
| Age group | 18–19 | 17 | 5.1 |
| 20–29 | 61 | 18.2 | |
| 30–39 | 87 | 25.9 | |
| 40–49 | 60 | 17.9 | |
| 50–59 | 79 | 23.5 | |
| 60+ | 32 | 9.5 | |
| Education | G7/No education | 11 | 3.3 |
| G10 | 46 | 13.7 | |
| G12 | 167 | 49.7 | |
| College diploma | 77 | 22.9 | |
| University degree | 35 | 10.4 | |
| Importance of religion | Not at all important | 12 | 3.6 |
| Not too important | 25 | 7.4 | |
| Somewhat important | 75 | 22.3 | |
| Very important | 224 | 66.7 | |
| Home language | Afrikaans | 112 | 33.3 |
| Bantu language | 138 | 41.1 | |
| English | 80 | 23.8 | |
| Other | 6 | 1.8 |
On the main question, 73.7% (n = 245) of respondents chose a human doctor, and 26.3% (n = 88) chose an AI doctor, p <.001. Pearson’s chi-square test was used to investigate whether there were any significant relationships between the demographic variables and the choice in the main question. Remarkably, gender, education, and home language did not have a significant relationship with the human/AI doctor preference (p =.858, 0.968 and 0.341 respectively). However, religion and age group did show such a relationship: A significant relationship was found between human/AI doctor preference and the importance of religion, χ2 (3) = 17.782, p <.001. A significant proportion of those for whom religion is “somewhat important” indicated that they would prefer a human doctor, as opposed to an AI doctor, while a significant proportion of those for whom religion is “not too important” would prefer an AI doctor as opposed to a human doctor. Notably, however, in the two groups at the extreme poles of the importance-of-religion question, namely “very important” and “not important”, no significant relationship with the human/AI doctor preference exists.
Also, there is a significant relationship between human/AI doctor preference and age, χ2 (5) = 12.831, p =.025. A significant proportion of 40–49 year olds indicated that they would prefer an AI doctor, as opposed to a human doctor.
Discussion
The central question of this study was whether South Africans would prefer to trust a human doctor or an AI doctor. The findings reveal that a significant majority (73.7% after adjustment) expressed a preference for human doctors over AI doctors. This result aligns with global trends reported in the literature, where AI is recognised for its potential to enhance healthcare outcomes, yet public trust largely remains anchored in the human element of healthcare decision-making.
The observed preference for human doctors underscores—at least for now—the importance of maintaining the human touch in healthcare, even as AI technologies advance. Notably, in this sample, gender, education, and home language did not show a significant relationship with human/AI doctor preferences. This finding is consistent with some studies but contrasts with others, highlighting the contextual specificity of these dynamics [24, 27, 33, 38, 47]. That education level does not influence AI trust contradicts Tyson et al. [24] and suggests that factors other than formal education, such as cultural familiarity with technology, may play a greater role in shaping public perceptions.
The influence of religion and age on preferences reveals more nuanced dynamics. The observed attenuation of the effect of religious importance at its extremes suggests that trust in AI is influenced by a complex interplay of belief systems and perceived morality of technological interventions. In this context, it is relevant to note that over 85% of South Africans identify as Christian, while traditional African religion is followed by nearly 8%, and just under 3% report no religious affiliation [49]. The openness to AI among individuals aged 40–49, a group not typically associated with technological enthusiasm, is particularly intriguing. This finding contrasts with international studies [24, 27, 38, 47] and points to unique generational or socio-economic factors in the South African context, warranting further investigation.
The survey results should not be interpreted as a wholesale rejection of AI-driven healthcare without human oversight. Instead, it highlights the need to understand public trust dynamics and address specific concerns that shape these preferences.
Limitations and future directions
This study has various limitations, notably the reliance on online recruitment and self-reporting, which may introduce biases. While the use of Facebook for recruitment might exclude individuals at the margins of digital access, its reach in South Africa is substantial, as discussed above. Accordingly, any exclusion introduced by the recruitment method is likely minimal.
Additionally, the simplification of the choice between a human doctor and an AI doctor does not capture the complexities of AI-human collaboration in healthcare. Future research should explore these collaborative models in more detail, examining how AI can augment rather than replace human medical expertise. The influence of religion and age on trust in AI also suggests the need for deeper investigation into how cultural, ethical, and social factors shape public attitudes toward AI in healthcare. Qualitative studies could provide richer insights into the concerns underpinning public trust, while longitudinal studies would offer valuable perspectives on how these perceptions evolve over time.
A less frequently mentioned demographic measure featured in the literature is political conservatism. We chose not to include it here as it is not a descriptor often encountered in South Africa. However, future studies may benefit from exploring political leaning or affiliation and its relevance within the South African setting.
While this study is preliminary, it provides an empirical starting point for understanding public perceptions of AI in South African healthcare. The strong preference for human doctors highlights public caution about AI autonomy, particularly in high-stakes healthcare decisions. This finding underscores the importance of addressing these apprehensions through further research and dialogue.
In conclusion, the results of this study indicate that a significant majority of South Africans currently prefer human doctors over AI doctors in the context of serious illness. This highlights the prevailing trust in the human element of healthcare decision-making. As AI technologies continue to develop, further research will be needed to understand whether and how societal attitudes may shift over time and under what conditions AI could gain greater acceptance in healthcare settings.
Supplementary Information
Below is the link to the electronic supplementary material.
Acknowledgements
The authors acknowledge the assistance of Gill Hendry with the statistical analysis, and of ChatGPT4 in summarising content, and in enhancing the language and readability of this manuscript.
Author contributions
DT – conceptualisation, project co-ordination, writing of original draft, revision, funding acquisition; DB – writing of original draft, revision.
Funding
The HSRC/Facebook Ethics & Human Rights and AI in Africa grant. The US National Institute of Mental Health and the US National Institutes of Health (award number U01MH127690). The content of this article is solely the authors’ responsibility and does not necessarily represent the official views of the two respective funders.
Data availability
Data is provided in supplementary information files.
Declarations
Ethics approval and consent to participate
Ethics approval was granted by the University of KwaZulu-Natal’s Biomedical Research Ethics Committee. Reference number: 6176/2023.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Clinical trial number
Not applicable.
Footnotes
The original online version of this article was revised: all the reference citations are hyperlinked.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Change history
9/15/2025
All the reference citations are hyperlinked.
Change history
9/30/2025
A Correction to this paper has been published: 10.1186/s12910-025-01293-3
References
- 1.Poalelungi DG, Musat CL, Fulga A, Neagu M, Neagu AI, Piraianu AI, Fulga I. Advancing patient care: how artificial intelligence is transforming healthcare. J Pers Med. 2023;13(8):1214. 10.3390/jpm13081214. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Qayyum MU, Sherani AMK, Khan M, Hussain HK. Revolutionizing healthcare: the transformative impact of artificial intelligence in medicine. Bull Inform. 2024;1(2):71–83
- 3.Sezgin E. Artificial intelligence in healthcare: complementing, not replacing, doctors and healthcare providers. Digit Health. 2023. 10.1177/20552076231186520. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Olawade DB, Wada OJ, David-Olawade AC, Kunonga E, Abaire O, Ling J. Using artificial intelligence to improve public health: a narrative review. Front Public Health. 2023;11:1196397. 10.3389/fpubh.2023.1196397. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Bhattamisra SK, Banerjee P, Gupta P, Mayuren J, Patra S, Candasamy M. Artificial intelligence in pharmaceutical and healthcare research. Big Data Cogn Comput. 2023. 10.3390/bdcc7010010. [Google Scholar]
- 6.Mbunge E, Batani J. Application of deep learning and machine learning models to improve healthcare in sub-Saharan Africa: emerging opportunities, trends and implications. Telemat Inf Rep. 2023;11: 100097. 10.1016/j.teler.2023.100097. [Google Scholar]
- 7.Oladipo EK, Adeyemo SF, Oluwasanya GJ, Oyinloye OR, Oyeyiola OH, Akinrinmade ID, et al. Impact and challenges of artificial intelligence integration in the African health sector: a review. Trends Med Res. 2024;19(1):220–35. 10.3923/tmr.2024.220.235. [Google Scholar]
- 8.Owoyemi A, Owoyemi J, Osiyemi A, Boyd A. Artificial intelligence for healthcare in Africa. Front Digit Health. 2020;2:6. 10.3389/fdgth.2020.00006. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Choudhury A, Elkefi S. Acceptance, initial trust formation, and human biases in artificial intelligence: focus on clinicians. Front Digit Health. 2022;4:966174. https://www.frontiersin.org/articles/10.3389/fdgth.2022.966174/full. Accessed 25 August 2024. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Shamszare H, Choudhury A. Clinicians’ perceptions of artificial intelligence: focus on workload, risk, trust, clinical decision making, and clinical integration. Healthcare (Basel). 2023. 10.3390/healthcare11162308. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Chandio SA, Rehman AU, Bano S, Hammed A, Hussain A. Enhancing trust in healthcare: the role of AI explainability and professional familiarity. Asian Bull Big Data Manag. 2024;4(1):12–21. 10.62019/abbdm.v4i1.100. [Google Scholar]
- 12.Kerasidou A, Kerasidou C, Buscher M, Wilkinson S. Before and beyond trust: reliance in medical AI. J Med Ethics. 2022;48(11):852–6. 10.1136/medethics-2020-107095. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Gille F, Jobin A, Ienca M. What we talk about when we talk about trust: theory of trust for AI in healthcare. Intell Based Med. 2020;1:100001. 10.1016/j.ibmed.2020.100001.
- 14.Robles P, Mallinson DJ. Artificial intelligence technology, public trust, and effective governance. Rev Policy Res. 2025;42(1):11–28. 10.1111/ropr.12555. [Google Scholar]
- 15.Mayosi BM, Benatar SR. Health and health care in South Africa–20 years after Mandela. N Engl J Med. 2014;371(14):1344–53. 10.1056/NEJMsr1405012. [DOI] [PubMed] [Google Scholar]
- 16.Harris B, Goudge J, Ataguba J, et al. Inequities in access to health care in South Africa. J Public Health Policy. 2011;32(Suppl 1):S102–23. 10.1057/jphp.2011.35. [DOI] [PubMed] [Google Scholar]
- 17.Amann J, Vayena E, Ormond KE, Frey D, Madai VI, Blasimme A. Expectations and attitudes towards medical artificial intelligence: a qualitative study in the field of stroke. PLoS ONE. 2023;18(1):e0279088. 10.1371/journal.pone.0279088. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Turchioe MR, Harkins S, Desai P, Kumar S, Kim J, Hermann A, et al. Women’s perspectives on the use of artificial intelligence (AI)-based technologies in mental healthcare. JAMIA Open. 2023;6(3):ooad048. 10.1093/jamiaopen/ooad048. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Dousa R. Toward the clinic: understanding patient perspectives on AI and data-sharing for AI-driven oncology drug development. In: Cassidy JW, Taylor B, editors. Artificial intelligence in oncology drug discovery and development. IntechOpen; 2020. 10.5772/intechopen.92787
- 20.Scantamburlo T, Cortés A, Foffano F, Barrué C, Distefano V, Pham L, et al. Artificial intelligence across Europe: a study on awareness, attitude and trust. arXiv [preprint]. 2023. https://arxiv.org/abs/2308.09979v1 Accessed 25 Aug 2024.
- 21.Beets B, Newmann TP, Howell EL, Bao L, Yang S. Surveying public perceptions of artificial intelligence in health care in the United States: systematic review. J Med Internet Res. 2023;25:e40337. 10.2196/40337. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Khullar D, Casalino LP, Qian Y, Lu Y, Krumholz HM, Aneja S. Perspectives of patients about artificial intelligence in healthcare. JAMA Netw Open. 2022;5(5):e2210309. 10.1001/jamanetworkopen.2022.10309. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Richardson JP, Smith C, Curtis S, Watson S, Zhu X, Barry B, et al. Patient apprehensions about the use of artificial intelligence in healthcare. NPJ Digit Med. 2021;4: 140. 10.1038/s41746-021-00509-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Tyson A, Pasquini G, Spencer A, Funk C, PEW Research Center. 60% of Americans would rather be uncomfortable with provider relying on AI in their own health care. 2023. https://www.pewresearch.org/science/2023/02/22/60-of-americans-would-be-uncomfortable-with-provider-relying-on-ai-in-their-own-health-care/ Accessed 25 Aug 2024.
- 25.Couture V, Roy M-C, Dez E, Laperle S, Bélisle-Pipon J-C. Ethical implications of artificial intelligence in population health and the public’s role in its governance: perspectives from a citizen and expert panel. J Med Internet Res. 2023;25:e44357. 10.2196/44357. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Hemphill S, Jackson K, Bradley S, Bhartia B. The implementation of artificial intelligence in radiology: a narrative review of patient perspectives. Future Healthc J. 2023;10(1):63–8. 10.7861/fhj.2022-0097. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Robertson C, Woods A, Bergstrand K, Findley J, Balser C, Slepian MJ. Diverse patients’ attitudes towards artificial intelligence (AI) in diagnosis. PLoS Digit Health. 2023;2(5):e0000237. 10.1371/journal.pdig.0000237. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Darcel K, Upshaw T, Craig-Neil A, Macklin J, Gray CS, Chan TCY, Gibson J, Pinto AD. Implementing artificial intelligence in Canadian primary care: barriers and strategies identified through a national deliberative dialogue. PLoS ONE. 2023;18(2):e0281733. 10.1371/journal.pone.0281733. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Gong W. From prostate cancer patient perspective: artificial intelligence may be a damn good idea [Master’s thesis]. Harvard University. 2023. https://nrs.harvard.edu/URN-3:HUL.INSTREPOS:37373934. Accessed 25 Aug 2024.
- 30.Carter SM, Carolan L, Aquino YSJ, Frazer H, Rogers WA, Hall J. Australian women’s judgements about using artificial intelligence to read mammograms in breast cancer screening. Digit Health. 2023;9:1–14. 10.1177/20552076231191057. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Esmaeilzadeh P, Mirzaei T, Dharanikota S. Patients’ perceptions toward human-artificial intelligence interaction in health care: experimental study. J Med Internet Res. 2021;23(11):e25856. 10.2196/25856. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Nussberger A-M, Luo L, Crockett MJ. Public attitudes value interpretability but prioritize accuracy in artificial intelligence. Nat Commun. 2022;13:5821. 10.1038/s41467-022-33417-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Young AT, Amara D, Bhattacharya A, Wei ML. Patient and general public attitudes towards clinical artificial intelligence: a mixed methods systematic review. Lancet Digit Health. 2021;3:e599–611. 10.1016/S2589-7500(21)00132-1. [DOI] [PubMed] [Google Scholar]
- 34.Nasim SF, Rizwan A, Kulsoom U. Artificial intelligence incidents & ethics a narrative review. Int J TIM. 2022. 10.54489/ijtim.v2i2.80. [Google Scholar]
- 35.Temple S, Rowbottom C, Simpson J. Patient views on the implementation of artificial intelligence radiotherapy. Radiography (London). 2023;29:S112–6. 10.1016/j.radi.2023.03.006. [DOI] [PubMed] [Google Scholar]
- 36.Lennartz S, Dratsch T, Zopfs D, Persigehl T, Maintz D, Hokamp NG, et al. Use and control of artificial intelligence in patients across the medical workflow: single-center questionnaire study of patient perspectives. J Med Internet Res. 2021;23:e24221. 10.2196/24221. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Thai K, Tsiandoulas KH, Stephenson EA, Menna-Dack D, Shaul RZ, Anderson JA, et al. Perspectives of youths on the ethical use of artificial intelligence in health care research and clinical care. JAMA Netw Open. 2023;6(5):e2310659. 10.1001/jamanetworkopen.2023.10659. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Parry MW, Markowitz JS, Nordberg CM, Patel A, Bronson WH, DelSole EM. Patient perspectives on artificial intelligence in healthcare decision making: a multi-center comparative study. Indian J Orthop. 2023;57:653. 10.1007/s43465-023-00845-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Elendu C, Amaechi DC, Elendu TC, Jingwa KA, Okoye OK, Okah MJ, Ladele JA, Farah AH, Alimi HA. Ethical implications of AI and robotics in healthcare: a review. Medicine (Baltimore). 2023;102(50):e36671. 10.1097/MD.0000000000036671. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Ferlito B, Segers S, De Proost M, Mertes H. Responsibility gap (s) due to the introduction of AI in healthcare: an Ubuntu-inspired approach. Sci Eng Ethics. 2024;30(4):34. 10.1007/s11948-024-00501-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Naidoo S, Bottomley D, Naidoo M, Donnelly D, Thaldar DW. Artificial intelligence in healthcare: proposals for policy development in South Africa. S Afr J Bioeth Law. 2022;15(1):11–6. 10.7196/SAJBL.2022.v15i1.797. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Sihlahla I, Donnelly D-L, Townsend B, Thaldar D. Legal and ethical principles governing the use of artificial intelligence in radiology services in South Africa. Dev World Bioeth. 2023. 10.1111/dewb.12436. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43.Townsend BA, Sihlahla I, Naidoo M, Naidoo S, Donnelly D-L, Thaldar DW. Mapping the regulatory landscape of AI in healthcare in Africa. Front Pharmacol. 2023;14:1214422. 10.3389/fphar.2023.1214422. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44.Eke DO, Wakunuma K, Akintoye S, editors. Responsible AI in Africa: challenges and opportunities. Cham: Springer Nature; 2023. Available from: 10.1007/978-3-031-08215-3
- 45.Kabata F, Thaldar D. Human in the loop requirement and AI healthcare applications in low-resource settings: a narrative review. S Afr J Bioeth Law. 2024;17(2):e1975. 10.7196/SAJBL.2024.v17i2.1975. [Google Scholar]
- 46.DataReportal, Meltwater, We Are Social. Digital 2024: global overview report. DataReportal; 2024. Available from: https://datareportal.com/reports/digital-2024-global-overview-report
- 47.York T, Jenney H, Jones G. Clinician and computer: a study on patient perceptions of artificial intelligence in skeletal radiology. BMJ Health Care Inform. 2020;27:e100233. 10.1136/bmjhci-2020-100233. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48.Obasa AE. The ethics of artificial intelligence in healthcare settings. Stellenbosch University, Stellenbosch, South Africa. 2023. Available from: https://scholar.sun.ac.za/server/api/core/bitstreams/b96f84c8-467f-41f4-b4fa-68fd8872b800/content
- 49.Statistics South Africa. Census 2022: statistical release P0301.4. Pretoria: Statistics South Africa; 2023. Available from: https://census.statssa.gov.za/assets/documents/2022/P03014_Census_2022_Statistical_Release.pdf
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Data is provided in supplementary information files.