Advances in data science and medical artificial intelligence (AI) raise complex philosophical and ethical quandaries about what it means to know a person or a community through data, and what kinds of people and societies we are becoming in this era of predictive data science, write N S Munung and G B Tangwa in the SA Medical Journal.
Using four lightly fictional but reality-informed case studies in mental health, radiology, genomics and environmental public health, they reflect on how AI technologies, largely built on Western biomedical traditions, may conflict with relational, spiritual and indigenous understandings of health and well-being.
Munung and Tangwa write:
This may manifest as epistemic friction, algorithmic fatalism and diminished trust in patient–clinician relationships, but besides familiar concerns regarding bias and transparency, this paper advances the discourse on the ethics of medical AI and data science in healthcare by shifting the analysis from epistemology (how AI systems know, classify and predict) to ontology (the study of the nature of being, as reconfigured by data and AI).
The future of medical AI must be defined by the moral and philosophical traditions by which people already live, and wherever they are; their world views should not be treated as peripheral critiques, but as constitutive resources for building inclusive, human-centred health technologies; while bioethics should be recognised as a core infrastructure in global health, on equal footing with data science, medicine, biomedical research and health innovation.
As data science and medical AI tools are increasingly deployed, they carry not only code and technical capabilities but ontological assumptions about health, well-being and personhood. These tools enter healthcare and social spaces as non-neutral artefacts.
Most also embody preconceptions shaped by Western traditions of risk quantification, biomedical abstraction and individualism, influencing how health and well-being are conceived, defined, interpreted and governed.
The rendering of granular personal, biological and environmental data into machine-readable form, known as ‘datafication’ and ‘datafied embodiment’, raises philosophical and ethical questions about which forms of knowledge are recognised, privileged or marginalised within algorithmic systems.
Where structural determinants of health intertwine with communal relations, lands, spirituality, cultural traditions, environmental conditions and human-non-human interactions, reducing complex experiences to discrete data points may result in epistemically partial representations and tension with local conceptions of being, health and well-being.
Across Africa, medical AI adoption is relatively nascent, however, investment, innovation and implementation efforts are accelerating rapidly.
Some pilots include AI-assisted diagnosis of diabetic retinopathy in Zambia, rapid automated TB screening in hard-to-reach regions in Uganda, and mental health chatbots for students and adolescents in South Africa and Cameroon.
When AI is embedded within clinical workflows, public health surveillance and disease prediction, it inevitably interacts with local moral frameworks, cultures and everyday practices.
However, the design of many such tools is informed by Western biomedical models that prioritise individualised risk assessment, statistical inference and predictive optimisation, even where health, illness and well-being are understood relationally, spiritually or ecologically.
This creates the risk of epistemic disjuncture between algorithmic outputs and local knowledges of being, and raises the question: whose knowledge counts in medical AI design, deployment and governance?
What does it mean to ‘know’ a person or a community through predictive scores, digital biomarkers and probabilistic inference? How can agency, trust and informed consent be preserved in healthcare when AI reasoning remains partially opaque, even to clinicians?
Rather than treating ethical and epistemic concerns as secondary to the technical performance of medical AI, we must foreground its ontological and normative assumptions and examine how they align – or fail to align – with diverse moral worldviews of health, wellness, personhood and the meaning of life.
As a methodological device for ethical analysis, we used four fictional case studies in mental health, radiology, genomics and environmental public health to reflect on how data science and medical AI may generate ethical tensions around epistemic injustice, trust, agency and responsibility.
First, we show that the ethical challenges of AI in medicine span ontology, epistemology and moral agency. Second, that medical humanism, enriched by inter-philosophies dialogue and cross-cultural ethics, is necessary for AI governance in global health. And third, that ecocentric ethics is necessary for addressing the relational and environmental dimensions of health routinely overlooked by AI systems.
We ask how such systems reconfigure what it means to be a person, a patient or a community, and offer a complementary normative lens for AI ethics. We conclude by arguing that bioethics should be treated as a key infrastructure for emerging technologies, on par with data science, medicine, biomedical research and innovation.
Case studies
Case 1:
AI-powered mental health assessments, epistemic friction and ontological harm
In a pilot digital mental health programme in a school in Bamenda, Cameroon, a UK-based start-up deployed SweetClassMH, an AI-powered screening chatbot, to support early identification of emotional distress among students, who were encouraged to use it as part of routine mental health monitoring.
During one interaction, the system classified Sirri, (15) and grieving the recent death of her twin sister, as ‘high risk’ for a mental health problem, recommending counselling and psychotherapy.
Sirri interpreted this to mean she was ‘going mad’. She became anxious, withdrew socially, told the school counsellor her family considered her present condition to be twin-soul bereavement, and that all she needed was to keep with her traditional symbol of her twin for spiritual stability.
In her culture, twins are considered to be one soul in two bodies.
In this case, the algorithmic classification of mental health clashes with the moral and spiritual frameworks held by Sirri’s family. It illustrates how a well-intentioned AI mental health application can generate epistemic friction in contexts where mental health or psychological suffering is not primarily understood through biomedical categories.
The ethical concern lies in algorithmic authority displacing spiritual, relational or cultural understandings of emotional stability. Can emotional distress be meaningfully reduced to data patterns?
More fundamentally, SweetClassMH inflicted ontological harm, understood here as damage to a person’s sense of self via imposed classificatory regimes from another world view. For Sirri, this involved reconfiguring her identity from a grieving twin sister who has lost a part of her ‘soul’ to a person perceived as mentally ‘abnormal’.
The normative issue, therefore, is not whether the diagnosis from SweetClassMH was accurate, but whose conception of mental health informed its design and deployment.
A growing body of literature suggests that conversational AI in mental health can misinterpret culturally situated expressions of distress, reinforce stigma or pathologise ordinary emotional responses when local worldviews are ignored.
This case raises three questions. First, whose definitions of well-being should guide AI mental health tools?
Second, how might such systems accommodate plural and sometimes conflicting understandings of mental health and its causes without reducing one to another? Third, what forms of accountability are owed, and by whom (in this case, the school or the developers), for ontological harm, even in the absence of clinical error?
These are not questions only for data scientists or psychologists, but require engagement with philosophers of mind, anthropologists, educators, spiritual leaders, bioethicists and ordinary non-experts.
Case 2:
The black box of AI in radiology and the philosophical limits of knowing AI systems are increasingly integrated into radiology to support the detection and classification of disease using medical imaging like mammograms, magnetic resonance imaging (MRI) and computed tomography (CT) scans.
These often rely on inscrutable machine-learning models inaccessible to clinicians and patients.
In an SA public hospital in Mthatha, Eastern Cape, a multinational health technology firm has deployed an AI-powered mammography system for breast cancer screening. It was trained on imaging data sets from women in Kenya, Nigeria, Tunisia and SA.
A mammogram for Nomfundo, a 42-year-old teacher, was initially interpreted by a radiologist as normal. The AI system, however, classified it as ‘high risk’, recommending a biopsy.
When Nomfundo asked why, the clinician responded: ‘It detected a pattern we do not fully understand, but is based on data from women across Africa.’ Nomfundo consented to the biopsy out of fear of delaying care rather than confidence in the decision.
Although the biopsy confirmed early-stage breast cancer, she was left anxious and distrustful of the diagnostic process.
This encounter exemplifies the philosophical limits of knowing in AI-assisted care, and how opacity in AI-assisted diagnosis can undermine patient agency even when outcomes are clinically beneficial.
The ‘black box’ logic displaces traditional trust in the patient–clinician relationship, namely empathy, dialogue and explanation of diagnosis and care pathways. What does it mean for clinical judgment when disease is identified through data patterns that are epistemically inaccessible to both patient and clinician?
The algorithm may have been correct, but it disrupted the healthcare experience.
This case also raises questions of accountability and professional responsibility. Had the biopsy revealed no cancer, it would have been difficult to assign responsibility for the physical risk and emotional stress to which Nomfundo was exposed. Conversely, had AI not been used and the cancer been missed, clinicians might have been faulted for not using available tools.
AI thus introduces a terrain of diffused responsibility and blurs established norms of clinical accountability.
While data representation in AI training is ethically, scientifically important, representation in data sets does not equate to representation in epistemic values or philosophies of healthcare.
Epistemic justice in medical AI demands interpretability plus participatory engagement with local practices, expectations and ontologies of care. Otherwise, technically sound clinical decisions may estrange patients ontologically.
Case 3:
Genomic prediction, AI and narratives of fate AI-driven genomic technologies are increasingly offered through direct-to-consumer genetic testing platforms using machine-learning models to forecast individual risk for conditions like Alzheimer’s, cancer, diabetes and selected behavioural traits.[23]
Advertised as predictive, personalised and empowering, these tools promise to enable individuals to anticipate and manage their future health trajectories.
Danilo (28), of Afro-Brazilian heritage living in Toronto, Canada, ordered one such genomic testing kit out of curiosity rather than medical necessity. The AI-generated report assigned him a 70% lifetime risk of developing Alzheimer’s disease.
Although framed as probabilistic, the result was experienced as quasi-decisive. His parents interpreted the result through different epistemic frameworks.
Danilo and his mother practise a diasporic spiritual tradition and understood the result as ancestral fate or karma, while his father, a clinician and scientist, urged immediate preventive intervention. Confronted with these competing interpretations, Danilo began to reorganise his life.
He reconsidered fatherhood, revised his career plans and adopted new dietary and exercise regimens, not as expressions of empowerment but as strategies for controlling his future. What was presented as statistical risk became a normative directive.
This case illustrates how AI-enabled genomic prediction does more than merely inform future healthcare decisions. Here, three epistemologies intersect: techno-scientific, spiritual and clinical, each offering distinct accounts of what it means to know, interpret and act upon the future.
In moral traditions where the future is unknowable and embedded in spiritual narratives, algorithmic prediction risks imposing a deterministic script. The question, therefore, is: does knowing our statistical or probabilistic futures free or bind us?
In this case, the ethical concern is algorithmic fatalism: the belief that probabilistic genomic information determines destiny and narrows the space for meaningful choice. When individuals begin to live as if a predicted outcome were inevitable, genomic risk prediction becomes a technology of identity formation.
Who should determine how individuals live in light of probabilistic genomic information? What responsibilities do developers, clinicians and regulators bear in defining how risk is framed, interpreted and communicated in AI-enabled genomics without foreclosing the freedom to become otherwise?
To resist algorithmic fatalism, AI genomic tools must be embedded within frameworks that explicitly acknowledge the contingency of futures across different moral worlds. Inter-philosophies and cross-cultural dialogue are therefore required if predictive technologies are to support healthcare while preserving the freedom to become otherwise.
Case 4:
Environmental public health surveillance, AI and social injustice
Environmental public health surveillance is seeing growing demand for AI in outbreak detection, early warning, trend prediction and response modelling.
The Mpumalanga Highveld, a major industrial hub for coal mining and energy production in SA, is among the most polluted regions in the southern hemisphere. Decades of mining expansion have displaced families, destroyed sacred sites and desecrated ancestral graves.
In addition, chronic exposure to toxic air (‘black dust’) from the mines is associated with high rates of asthma, bronchitis and cardiovascular disease, and disproportionately affects black and low-income communities.
In 2023, a collaboration between an SA university and a Silicon Valley AI lab launched UmoyaScan, an environmental public health surveillance system that integrates satellite imagery, meteorological data, emissions modelling and self-reported health symptoms
to generate population-level risk profiles for pollution-related respiratory disease.
The Department of Health heralded UmoyaScan as a milestone in ‘precision public health’ in Africa. Yet for residents of the Highveld, this promise has not translated into meaningful health gains.
Doctors continue to report a high incidence of asthma, pneumonia and lung cancer, and local clinics are chronically under-resourced. Traditional healers, whose Indigenous knowledge systems have long tracked ecological disruption through land, wind patterns and animal behaviour, were not consulted in the design or implementation of UmoyaScan.
Some communities interpret the high number of lung cancer cases as a spiritual effect after the desecration of ancestral graves. Allegations have also emerged that community-level health and environmental data are being monetised in international climate and health finance markets without community consent or benefit sharing.
Although the government has acknowledged concerns regarding the health impact of ‘black dust’, data governance and environmental injustice, its political entanglements with the fossil fuel industry and national dependence on coal for energy have slowed remedial action.
UmoyaScan exemplifies how AI-enabled public health surveillance can reinforce structural, epistemic, environmental and ontological injustices rather than redress them. The system privileges algorithmic expertise and the generation of data for decision-making while neglecting local knowledge systems that guide the community responses to ecological harm.
Epistemically, UmoyaScan ‘sees’ pollution, particulate matter and pulmonary risk, but not moral injury and its association with community perceptions of prevalent respiratory conditions in the region.
Can satellite data capture the slow violence of industrial displacement? Who decides what counts as legitimate environmental health knowledge? Do communities have the moral right to refuse being mapped, predicted or monetised in the name of public health or climate action?
What UmoyaScan ultimately reveals is not a deficit of data, but a deficit of distributive, epistemic, ecological and spiritual justice.
A relational and ecocentric approach to AI governance in public health surveillance is needed, which centres community agency and interrogates who defines the goals, ownership and benefits of environmental AI systems.
Toward a medical humanism of inter-philosophies dialogue, cross-cultural ethics and ecocentricity
As AI gains influence in defining our medical futures, we must ask which philosophies of health, moral agency and knowing are being encoded, intentionally or unintentionally, into these systems.
Across the four cases presented, a consistent ethical pattern emerges: AI systems in medicine and public health do not merely assist clinical decision-making; they co-constitute how health, illness, personhood and responsibility are understood.
In Bamenda, algorithmic mental health classification transformed grief into pathology, generating fear and social withdrawal. In Mthatha, opaque algorithmic authority disrupted trust and agency despite diagnostic success.
In Toronto, probabilistic genomic prediction narrowed imagined futures and reoriented life choices toward anticipatory self-surveillance.
In Mpumalanga, environmental surveillance rendered communities legible as data sources while neglecting spiritual harm to land and culture.
These cases show that the ethical challenges of AI extend beyond epistemic failure to ontological harm. They concern philosophical questions about what it means to know, care for and live well as human beings situated within plural social, cultural and ecological worlds.
This clarifies why technical solutions like bias mitigation or XAI, while necessary, are insufficient on their own.
What is needed is a medical humanism placing AI within a pluralistic understanding of human flourishing, rather than reducing care to optimisation and prediction.
This is not a rejection of technology, but a call to maintain fundamental moral practices in medicine concerned with meaning, trustworthy patient–clinician relationships, dignity and relational responsibility. A medical humanism adequate to the age of AI must therefore attend not only to outcomes, but also to how technologies and data science reconfigure relationships, identities and agency.
Central to this humanistic reorientation are cross-cultural ethics, inter-philosophies dialogue and ecocentric perspectives.
This does not mean collapsing diverse traditions into a single framework, but creating bioethics spaces where tensions, complementarities and disagreements in medical AI can be openly discussed and negotiated.
In the age of AI, merely improving algorithms is not enough. We need to listen and act across disciplines, traditions and cultures.
Conclusion
This paper shifts contemporary discourse on AI ethics from predominantly epistemological concerns to the ontological effects on persons and communities. AI can generate significant ontological harm even when technically accurate.
Our proposal is to position AI within relational, plural and ecocentric visions of health, where humans are understood not merely as data sets or risk profiles but as meaning-making beings embedded in social, spiritual and ecological worlds.
As AI and data science become central and public to global health initiatives, a renewed social contract for data-driven medicine is required, which asks not only what AI can do, but what it ought to do, for whom, at what moral cost and to whose benefit.
What is needed is not only accurate or representative AI, but a medical humanism that recognises philosophy and bioethics as core infrastructures in global health, on equal footing with data science centres, clinical studies and biotechnology.
Otherwise, AI risks advancing as an extractive and epistemically narrow enterprise, one that draws on personal and environmental data while ignoring local meanings, values, injustice and histories.
N S Munung, MSc, PhD – Division of Human Genetics, Faculty of Health Sciences, University of Cape Town.
G B Tangwa, MA, PhD – Department of Philosophy, University of Yaoundé, Cameroon; Cameroon Bioethics Initiative (CAMBIN), Yaoundé, Cameroon
See more from MedicalBrief archives:
ChatGPT sued over health advice that discouraged medical care
Swift detection of cancer in mammograms by AI – Swedish study
AI can’t replicate this crucial aspect of practising medicine
OECD: How artificial intelligence could change the future of health
Lifesaving genetics tool works best if you’re white
South Africa forges ahead with tech-driven strategy to fight diseases