Artificial intelligence (AI) and medical robotics are no longer concepts confined to the realm of science fiction; they are actively reshaping the delivery of clinical care across South Africa. From advanced diagnostic imaging algorithms employed in private hospital groups to robotic-assisted surgical procedures and public sector pilot projects utilising triage chatbots, these innovative technologies are fundamentally transforming how critical medical decisions are made and how complex procedures are performed. This rapid technological integration, however, introduces a pressing and as yet unanswered question within the South African legal and ethical landscape: who bears legal responsibility when an AI system or a medical robot makes an error that results in patient harm?
This article delves into the burgeoning risks and liabilities intrinsically linked to the adoption and use of AI and robotics in the South African healthcare sector. It places a particular focus on the critical issues of institutional accountability, the imperative of informed patient consent, and the significant gaps that currently exist within the nation’s legal and regulatory frameworks. By drawing on insights from various sources, including case studies, academic commentary, and an examination of comparative international models, it becomes clear that South Africa is in urgent need of developing a robust and comprehensive governance model. Such a framework is essential to effectively keep pace with the accelerating realities of AI-powered and robot-assisted medicine, ensuring that innovation does not outstrip patient safety and legal clarity.
The Presence of AI and Robotics in South African Healthcare Today
The deployment of AI and robotic technologies in South African healthcare is increasingly widespread, encompassing a diverse array of applications across both the public and private sectors.
Where AI and Robotics are being used:
• Radiology and Pathology: AI algorithms are now routinely employed for the interpretation of medical imaging (like X-rays, CT scans, and MRIs) and histology slides. Examples include systems such as Aidoc and Zebra Medical, which assist clinicians in identifying abnormalities and making diagnoses with greater efficiency and precision.
• Clinical Decision Support: Sophisticated software tools are designed to recommend optimal treatment pathways for patients or to flag potentially dangerous drug interactions. These systems aim to enhance the quality and safety of clinical decisions by providing evidence-based guidance to healthcare professionals.
• Chatbots and Virtual Triage: Platforms, such as HealthChat, are being utilised in both urban and rural telehealth projects. These AI-powered chatbots can assist with initial patient screening, provide basic health information, and help in routing patients to appropriate care levels, particularly vital in under-resourced provinces.
• Administrative Tools: Beyond direct patient care, AI is also being leveraged to optimise administrative functions within healthcare institutions. This includes automating tasks such as scheduling appointments, efficiently allocating resources, and even predicting surges in patient admissions, thereby improving operational efficiency.
• Robotic Surgery: In high-end private hospitals, robotic surgical systems like the da Vinci Surgical System are actively used for a range of complex procedures, including urology, gynaecology, and general surgery. These systems offer enhanced precision, dexterity, and control for surgeons.
• Rehabilitation and Elderly Care: AI-powered robotic companions and advanced mobility aids are finding applications in frail care settings and for chronic patients, assisting with rehabilitation exercises and providing support for daily living activities.
Mixed Public-Private Adoption:
The adoption of these technologies exhibits a mixed pattern across South Africa’s healthcare landscape.
• Private Hospital Groups such as Mediclinic, Life, and Netcare are at the forefront, increasingly integrating both diagnostic support AI tools and advanced robotic surgical platforms into their services.
• Provincial eHealth strategies acknowledge and mention the role of AI in areas like disease surveillance and remote care. However, the use of robotics in the public sector remains significantly limited, primarily due to the substantial associated costs.
• The Department of Health’s Digital Health Strategy 2019–2024 references AI as a component of its vision but notably lacks a concrete legislative or regulatory framework specifically addressing the nuances of AI or robotics in clinical settings.
Comparative Legal and Ethical Risks: AI vs. Robotics
A useful way to understand the complexities introduced by these technologies is to compare the distinct legal and ethical risks they present.
| Dimension | AI | Robotics |
|---|---|---|
| Clinical Use | Diagnostics, decision support, triage | Surgery, rehabilitation, physical procedures |
| Risk of Harm | Misdiagnosis, bias, delayed care | Mechanical failure, haptic errors, procedural complications |
| Accountability Issues | “Black box” algorithms, unclear human oversight | Shared control between clinician and machine |
| Consent Challenges | Patients may not know AI was involved | Patients unaware robotics used in procedure |
| Regulatory Gap | No dedicated legislation or HPCSA guidance | No specific rules for robotic surgery or device classification |
| Case Law | None in South Africa | None in South Africa |
| Insurance Uncertainty | Not routinely covered in PI policies | Limited inclusion in surgical risk cover |
The Problem: When AI or Robotics Go Wrong
The integration of AI and robotics, while promising, is not without significant risks. When these technologies malfunction or provide flawed outputs, the consequences can be severe, leading to real harms for patients.
Real Risks, Real Harms:
• Misdiagnosis due to flawed AI output remains a primary concern. An incorrect interpretation by an algorithm can lead to delayed treatment or even death. For instance, an AI diagnostic tool might incorrectly identify a cancerous lesion as benign, leading to a critical delay in intervention.
• Biased algorithms pose a significant ethical and clinical challenge. If an AI model is trained predominantly on data from one demographic group, it may underperform or provide inaccurate results for certain races, ages, or genders. A stark example provided is an AI model trained primarily on Caucasian patient data potentially failing to accurately diagnose dermatological conditions in Black South Africans due to a lack of representative training data. This bias can exacerbate existing health inequities.
• Chatbots, particularly in remote care settings or under-resourced provinces, might downplay serious symptoms or misroute patients, preventing them from receiving timely and appropriate medical attention.
• Robotic surgery errors can have catastrophic physical consequences. These errors might stem from mechanical failure of the system, incorrect input programming by the human operator, or haptic malfunction (issues with the robot’s sense of touch and force feedback). Such errors can result in avoidable injury or even death during a procedure.
Who Takes the Blame? A Legal Quandary:
These real harms raise complex questions of accountability that remain largely unresolved under South African law.
• Is the hospital liable if it made the decision to procure, implement, and deploy the AI or robotic system?
• Is the software developer or the device manufacturer liable as a third-party vendor or supplier?
• What happens if the clinician relied on the AI’s recommendations or followed robotic guidance, but was subsequently overruled or constrained by the inherent design or limitations of the system?
These questions highlight a critical gap in the existing legal framework, creating a “regulatory grey zone” for AI and robotics in healthcare.
Legal Vacuum: Absence of Statutory or Case Law Guidance
South Africa’s current legal landscape is ill-equipped to address the complexities of liability stemming from AI and robotic errors in healthcare.
Common Law Unprepared:
• The established common law framework for delict (torts), which governs civil wrongs, relies on principles of negligence, wrongfulness, and causation. However, these principles were developed for human actors and human-centric medical practices.
• Crucially, no existing case law specifically addresses liability for AI tools or robotic systems in healthcare. This means there are no legal precedents to guide courts in attributing fault or responsibility in such cases.
• Current medico-legal principles assume human actors are at the centre of decision-making and procedural execution, making it difficult to apply them directly to autonomous or semi-autonomous systems.
Product Liability and the Consumer Protection Act (CPA): Limited Application:
• AI tools and robotic systems may potentially qualify as “products” under the Consumer Protection Act (CPA) 68 of 2008. If they do, the supplier could be held strictly liable for harm caused by a defect in the product.
• However, Section 61 of the CPA only applies where harm arises directly from a “defect”. The challenge arises because harm from AI often stems not from a manufacturing defect, but from “poor predictions” or flawed algorithmic outputs, which may not readily fit the definition of a “defect” in the traditional sense.
• This creates practical difficulties; for example, a private hospital in Johannesburg using an AI diagnostic tool supplied by a foreign developer might struggle to shift blame to the developer under current local laws due to the nuanced nature of AI errors. The “black box” nature of some AI models further complicates proving a defect versus a predictive error.
Absence of HPCSA or Statutory Guidelines:
• The Health Professions Council of South Africa (HPCSA), the professional body regulating healthcare practitioners, has not yet issued specific guidance on the use of AI decision-support tools or robotic systems. This lack of professional guidelines leaves clinicians and institutions without clear standards of practice regarding these technologies.
• Similarly, the Medicines and Related Substances Act does not currently cover AI-specific software or robotic devices unless they are classified as medical devices within the existing regulatory framework, which often doesn’t fully encompass the unique nature of software as a medical device (SaMD) or autonomous systems.
Institutional Liability: What Happens to Hospitals and Clinics?
As the primary deployers of these advanced technologies, hospitals and clinics face significant questions regarding their potential liability.
Vicarious Liability Revisited:
If a hospital actively deploys AI to guide clinical decisions or robotic systems for procedures, it may be found vicariously liable for any resulting harm. This legal concept holds an employer responsible for the actions of its employees. However, applying it to AI and robots introduces new complexities:
• Control: A key question becomes who procured and ultimately implemented the AI or robotic system within the institution. The extent of the hospital’s control over the technology’s selection, configuration, and integration is crucial.
• Oversight: Was there an adequate “human-in-the-loop” safeguard? This refers to whether sufficient human oversight and intervention mechanisms were in place to review, validate, and potentially override AI recommendations or robotic actions. Without such oversight, the hospital’s responsibility could increase.
• Consent: Critically, were patients fully informed and did they consent to the involvement of AI or robotics in their care? Lack of informed consent can be an independent ground for liability.
The “Black Box” Dilemma:
• A significant challenge for accountability lies in the “black box” nature of many AI tools. These systems often operate using complex algorithms that are not fully explainable or transparent, making it exceedingly difficult to understand why a particular decision or prediction was made.
• This opacity complicates the task of disproving negligence. If a hospital cannot explain how an AI arrived at a flawed diagnosis, it becomes harder to demonstrate that it exercised reasonable care in deploying and relying on that system.
• South African courts may soon be forced to consider whether the very act of deploying unexplainable algorithms in clinical settings violates the fundamental duty of care owed to patients. This could set a precedent requiring greater transparency in AI systems used in healthcare.
Informed Consent: Are Patients Told AI or Robotics Are Involved?
The principle of informed consent is a cornerstone of patient rights and ethical medical practice. However, its application in the context of AI and robotics is currently underdeveloped.
Current Law:
• The National Health Act 61 of 2003 and existing HPCSA guidelines mandate that patients be fully informed about the nature and purpose of their proposed treatment, its associated risks and benefits, and any available alternative options.
• However, there is no specific legal or regulatory requirement to disclose the use of AI or robotics in a patient’s diagnosis or treatment process.
Ethical Implications:
The absence of explicit disclosure requirements severely undermines the very essence of informed consent. Patients cannot make truly autonomous decisions if they are unaware of critical aspects of their care, such as:
• That an algorithm, rather than solely a human doctor, was responsible for flagging their condition or recommending a course of action. This can impact patient trust and perception of care.
• That the AI model they are being subjected to was trained on foreign data, which might not be representative of the South African population’s specific health profiles or genetic predispositions. This links back to the risk of biased algorithms.
• That their own medical data may be used to train future AI systems, raising privacy and data governance concerns.
A poignant example of this ethical gap is highlighted in a Cape Town case currently under academic review, where a patient was reportedly not informed that an AI tool had flagged a misdiagnosed lesion. This lack of disclosure significantly contributed to a delayed biopsy, potentially impacting the patient’s prognosis.
Comparative Models: What Other Countries Are Doing
Globally, various jurisdictions are grappling with similar challenges and are beginning to develop regulatory frameworks to address AI and robotics in healthcare.
United Kingdom (UK) and European Union (EU):
• The UK General Medical Council (GMC) is actively engaged in developing AI-specific clinical guidance for healthcare professionals, aiming to provide clarity on responsible AI use.
• The EU AI Act (2024) represents a landmark legislative effort. It classifies medical AI as “high-risk,” imposing stringent requirements for transparency, oversight, human supervision, and risk management. This comprehensive approach aims to ensure safe and ethically sound AI deployment in critical sectors like healthcare.
United States of America (USA):
• The Food and Drug Administration (FDA) plays a crucial role in regulating Software as a Medical Device (SaMD), establishing guidelines for its pre-market approval and post-market surveillance.
• Significantly, case law is beginning to emerge in the USA specifically around negligence claims linked to clinical reliance on AI. One notable case involved a hospital being sued for relying on a flawed oncology AI tool. This case raised fundamental questions regarding the foreseeability of AI errors and the extent to which clinicians or institutions can rely on such systems without independent verification. These emerging precedents provide valuable insights for jurisdictions like South Africa.
What Should South Africa Do? A Legal and Ethical Roadmap
To navigate the complexities and harness the benefits of AI and robotics in healthcare responsibly, South Africa needs a proactive and multi-faceted approach encompassing law reform, ethical safeguards, and insurance considerations.
Law Reform:
• Amend the National Health Act and HPCSA rules to explicitly include AI-specific disclosures regarding their use in patient care. This would provide clarity and legal backing for informed consent processes.
• Clarify institutional and product liability in the context of AI and robotics. This might involve introducing new legislation or amending existing laws to address the unique challenges posed by these technologies, assigning clear lines of responsibility to hospitals, manufacturers, and clinicians.
• Mandate algorithmic audits for AI systems used in high-risk medical applications. Such audits would assess the performance, fairness, and safety of algorithms, reducing the “black box” problem and improving accountability.
Ethical Safeguards:
• Require AI explainability in diagnostic tools. Developers should be encouraged or mandated to design AI systems where their decision-making processes are understandable and auditable, allowing clinicians to verify and trust the output.
• Update consent forms to specifically disclose AI and robotic use. This ensures that patients are fully aware of all aspects of their treatment, empowering them to make truly informed decisions.
• Establish AI ethics boards within healthcare institutions. These boards would provide internal oversight, develop institutional policies, and address ethical dilemmas related to the adoption and use of AI and robotics.
Insurance and Risk Transfer:
• Expand professional indemnity and institutional policies to explicitly cover AI and robotic tools. Current insurance policies may not adequately account for these emerging risks.
• Review vendor contracts thoroughly to clarify liability between healthcare institutions and technology providers. Contracts should clearly delineate responsibilities in case of malfunction, error, or harm.
• Insurers should consider developing dedicated coverage for AI-related and robotic-related claims. As these technologies transition from pilot projects to widespread clinical use, the volume and complexity of claims are likely to increase, necessitating specialized insurance products.
Conclusion
South African healthcare stands on the precipice of an AI-driven transformation, offering unprecedented opportunities to enhance care delivery and efficiency. However, the current legal and ethical safeguards have demonstrably not kept pace with this rapid technological advancement. Healthcare institutions that embrace AI and robotic tools with the aim of improving patient care must simultaneously prepare to proactively manage the new and complex risks that these technologies inherently introduce.
Without clear guidance on informed consent, definitive regulatory clarity regarding liability, and robust institutional oversight mechanisms, the immense promise of AI and robotics in medicine could inadvertently give rise to new forms of patient harm and, consequently, new and challenging sources of legal liability. It is imperative for all stakeholders – clinicians, lawyers, policymakers, and insurers – to act decisively now to ensure that AI and robotics in healthcare truly enhance patient safety rather than eroding it. The time to construct a comprehensive governance framework for clinical AI and robotics is not in the aftermath of the first high-profile malpractice claim, but rather before the next major deployment goes live. Proactive legislative, ethical, and practical measures are essential to secure a future where technology responsibly serves the health of all South Africans.
