Artificial intelligence (AI) is rapidly transforming the field of medicine, offering unprecedented opportunities to improve healthcare delivery, diagnosis, and population health management. However, with its promise comes a risk of harm, particularly when AI systems are poorly designed, implemented without appropriate safeguards, or driven by commercial interests at the expense of public good. This paper explores what constitutes good and evil in medical AI, provides examples of both, and outlines ethical boundaries and practical steps to ensure that AI serves humanity.
AI in medicine refers to systems designed to assist with tasks such as diagnosis, prognosis, treatment recommendations, and public health surveillance. The good in medical AI lies in its capacity to enhance human well-being, reduce inequalities, and improve healthcare efficiency. AI applications can support clinical decisions, automate routine tasks, and extend healthcare reach to underserved populations (Rajkomar, Dean, & Kohane, 2019). Conversely, the potential for evil emerges when AI contributes to harm, reinforces inequities, or undermines essential human values such as compassion, accountability, and justice. This harm may arise from biased algorithms, opaque decision-making processes, or commercial exploitation that prioritises profit over patient welfare.
The Goods
One of the clearest demonstrations of AI’s positive contribution to medicine is in the field of early disease detection. AI systems trained on medical images have been shown to accurately detect conditions such as diabetic retinopathy and tuberculosis. A pivotal study demonstrated that an autonomous AI system could safely and effectively identify diabetic retinopathy in primary care settings, enabling earlier referrals and potentially preventing vision loss (Abràmoff, Lavin, Birch, Shah, & Folk, 2018). In tuberculosis screening, AI-based chest X-ray interpretation tools have been used in high-burden countries to prioritise patients for further diagnostic testing, particularly in settings where human expertise is limited (Codlin et al., 2025). These applications help address gaps in healthcare access and reduce delays in diagnosis and treatment.
AI has also supported public health surveillance, particularly during emergencies such as the COVID-19 pandemic. AI models combined data from health records, mobility patterns, and social media to predict outbreaks, identify hotspots, and inform targeted interventions. This contributed to more timely and effective public health responses and resource allocation (Bullock, Luccioni, Hoffmann, & Jeni, 2020).
The Evils
Despite these benefits, AI has also been linked to harms that can undermine trust and exacerbate health inequities. One of the most pressing concerns is algorithmic bias. AI systems trained on data that do not represent the diversity of patient populations may produce biased outcomes. For example, machine learning tools for dermatology developed primarily using images of lighter skin tones have been found to perform less accurately on darker skin. This can lead to missed or delayed diagnoses in patients from minority groups, reinforcing existing disparities (Adamson & Smith, 2018).
Commercial exploitation of AI is another area of concern. The rush to monetise AI in medicine has sometimes led to the deployment of systems that are insufficiently transparent or accountable. Proprietary algorithms may operate as black boxes, with their decision-making processes hidden from both clinicians and patients. This opacity undermines informed consent and shared decision-making, and can make it difficult to challenge or review AI-driven recommendations (Char, Shah, & Magnus, 2018).
Furthermore, there is a risk that excessive reliance on AI could erode the compassionate, human-centred aspects of healthcare. While AI can assist with routine tasks and reduce administrative burdens, it must not be seen as a replacement for human empathy and professional judgement. There is concern that as AI takes on a greater role, the patient-doctor relationship could become depersonalised, weakening one of the core foundations of medical practice (Panch, Szolovits, & Atun, 2019).
Ethical Boundaries for Responsible AI
To ensure that AI in medicine serves the common good rather than causes harm, clear ethical boundaries are needed. Transparency is essential. AI systems must be designed in ways that make their decision-making processes understandable and open to scrutiny. This is critical to maintaining trust, supporting informed consent, and enabling clinicians to integrate AI recommendations into their decision-making with confidence.
Fairness must also be prioritised. Developers need to ensure that AI tools are designed to promote equity rather than exacerbate disparities. This involves using diverse training datasets, actively auditing algorithms for bias, and engaging with communities to understand their needs and perspectives. Bias mitigation should be a central part of AI development and deployment, not an afterthought.
Accountability is another key principle. Developers, healthcare providers, and regulators share responsibility for ensuring that AI systems are safe, effective, and aligned with ethical principles. Regulatory frameworks should define standards for AI in healthcare and provide mechanisms for monitoring, evaluation, and redress when harm occurs (Char et al., 2018).
Compassion must remain central to healthcare, even as AI systems become more common. AI should be designed and used to support, rather than replace, the human connection between healthcare professionals and patients. The ultimate goal should be to free clinicians from administrative burdens and allow them to focus on what matters most: the well-being of the people they serve (Topol, 2019).
Towards Governance and Action
The development and use of medical AI should be guided by comprehensive national or regional governance frameworks that balance the promotion of innovation with the protection of public interest. Such frameworks need to address issues including data privacy, transparency, bias mitigation, and equitable access. They should be shaped through collaboration between governments, healthcare professionals, technologists, and civil society to ensure that they are both robust and responsive to local contexts and needs.
Education and capacity building are also essential. Healthcare professionals, public health experts, and policymakers must be equipped with the knowledge and skills needed to engage with AI critically and effectively. Training should address not only technical competencies but also the ethical, legal, and social implications of AI.
Finally, ongoing research is needed to evaluate the real-world impact of AI in healthcare. This research should assess not only clinical outcomes but also equity, patient safety, and the preservation of humanistic values. It should inform continuous improvement of AI systems and the policies that govern their use (Morley, Floridi, Kinsey, & Elhalal, 2020).
Conclusion
AI has the potential to greatly enhance healthcare, improving efficiency, accuracy, and access. However, without appropriate safeguards, it also carries the risk of causing harm, deepening inequities, and eroding core human values. The boundary between good and evil in medical AI lies in how these technologies are designed, implemented, and governed. By upholding principles of transparency, fairness, accountability, and compassion, and by embedding these principles in governance frameworks and professional practice, it is possible to ensure that AI serves as a tool for good in medicine.
References
Abràmoff, M. D., Lavin, P. T., Birch, M., Shah, N., & Folk, J. C. (2018). Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices. NPJ Digital Medicine, 1, 39.
Adamson, A. S., & Smith, A. (2018). Machine learning and health care disparities in dermatology. JAMA Dermatology, 154(11), 1247-1248.
Bullock, J., Luccioni, A., Hoffmann, P. H., & Jeni, L. A. (2020). Mapping the landscape of artificial intelligence applications against COVID-19. Journal of Artificial Intelligence Research, 69, 807-845.
Char, D. S., Shah, N. H., & Magnus, D. (2018). Implementing machine learning in health care – Addressing ethical challenges. New England Journal of Medicine, 378, 981-983.
Chen, I. Y., Szolovits, P., & Ghassemi, M. (2019). Can AI help reduce disparities in general medical and mental health care? AMA Journal of Ethics, 21(2), E167-E179.
Codlin, A. J., Dao, T. P., Vo, L. N. Q., Forse, R. J., Nadol, P., & Nguyen, V. N. (2025). Comparison of different Lunit INSIGHT CXR software versions when reading chest radiographs for tuberculosis. PLOS Digital Health, 4(4), e0000813.
Morley, J., Floridi, L., Kinsey, L., & Elhalal, A. (2020). From what to how: An overview of AI ethics tools, methods and research to translate principles into practices. AI & Society, 36, 59-71.
Panch, T., Szolovits, P., & Atun, R. (2019). Artificial intelligence, machine learning and health systems. Journal of Global Health, 8(2), 020303.
Rajkomar, A., Dean, J., & Kohane, I. (2019). Machine learning in medicine. New England Journal of Medicine, 380(14), 1347-1358.
Topol, E. (2019). Deep medicine: How artificial intelligence can make healthcare human again. Basic Books.