The only option: Closer to a chatbot than a clinic
Author: Prof Nasreen Jessani1
1. Centre for Evidence Based Health Care, Stellenbosch University, South Africa
‘I asked ChatGPT to help diagnose my problem.’ That was how my Zimbabwean friend Linda (not her real name) explained her solution to addressing four years of chronic back pain. Linda took this approach of asking artificial intelligence (AI) as her situation left little room for a costly diagnostic exam. She had no health insurance, was subject to xenophobic attacks, and sent much of what she earned home to family in Zimbabwe. Her own healthcare sat near the bottom of her priority list, and beyond the point on that list where she could afford to spend money finding a diagnosis.
So, ChatGPT was her first stop, and her only one. The diagnosis it offered set the course for almost four years: appointments scheduled, cancelled, rescheduled; scans; blood tests; rounds of antibiotics; and the quieter psychological cost of enduring an unresolved health problem that was never improved. Linda feared ovarian cancer, recurrent urinary tract infections, and failing kidneys. Nothing resolved her back pain, because the original diagnosis was wrong, and no one in the loop was positioned to say so.
What ended this cycle of wasted time and energy was unremarkable: a visit to a GP. After one conversation – where the GP asked about all of her symptoms, not just the few she had thought to type – and a physical exam, Linda came away with a different diagnosis, a treatment that worked, and the relief of finally knowing what the problem had been all along.
While this may be an extreme case, it’s a real example of the potential and perils of AI when it comes to medical information and diagnosis. We worry, rightly, about inequity when AI is out of reach of the poor – but Linda’s case is the opposite. Inequity in the health system, not in access to technology, pushed her toward a tool that was easier to reach than a clinic and far less able to help. This a clear example of how inequity in the health system endangers those who can access the internet more easily than they can book a health appointment. In this scenario, AI is not closing the gap to health equity; it is standing in for the necessary support they cannot reach.
That raises the question of this blog: how do we centre people who have high need, limited information literacy, and a strong, rational preference for solutions that are fast and free, and who are using AI to make medical decisions with no one accountable for the outcome?
This question is an important one to ask in the South African context. According to the 2025 Global Digital Report for South Africa, cell phone connections are estimated at close to 200% of the population, while online penetration sits at around 80% and social media use at 42%. Moreover, the South African Social Attitudes Survey (2023) indicated that 29% of people in rural areas and 74% of those in urban areas have regular access to the internet. These figures illustrate the proportion of people in South Africa with access to information at their fingertips. Now, with the advent of AI, curiosity, scepticism, and opportunity are all interwoven.
This phenomenon is not new. Discussions around the Fourth Industrial Revolution – of which AI is a technology of focus – have been floating around since 2019, when President Ramaphosa established the Presidential Commission on the Fourth Industrial Revolution to guide South Africa’s transition. The Commission reported in 2020, and its recommendations spoke of skills, infrastructure, and an enabling environment. Yet none of that reached Linda. What reached her instead was a free chatbot on a phone she already owned. While the Commission worried about the digital divide, we can see in 2026 that the divide closed faster than anyone planned, but not with entirely positive results for those gaining information through new digital technology.
This is the part that equity conversations keep getting backwards. We assume the danger is that AI will be hoarded by the well-resourced and withheld from the poor. Yet Linda’s four years suggest a different opportunity for harm. She had abundant access to AI and almost none to a clinician. The chatbot was not a supplement to her care; instead, it was her care, because the alternative was a system she could not afford to enter and safely move through. AI did not widen a gap here. It filled one – badly. And that outcome delayed Linda’s treatment by years.
That distinction matters for what we do next. If the problem is access to AI, then the answer is distribution. If the problem is that AI substitutes for a health encounter it cannot replicate, then distribution makes things worse, not better. Linda did not need a better prompt. Instead, she needed a 15-minute conversation with someone who could ask the right questions, draw from experience, and provide a pathway to health. The diagnosis that was ultimately correct arose through human interaction, which no model can deliver.
So, the ‘people at the centre’ reads differently in some areas of South Africa than it might to a research institute. The phrase usually points to involving patients in how models are built and tested, and that work matters. But Linda was not on the edge of a design process. Instead, she was on the edge of the health system itself, managing a clinical problem alone, with a tool that returned plausible sentences and owed her nothing. As this illustrates, efforts to keep people at the centre must start earlier than the model. These require identifying who has already been pushed out – that is, who uses AI not after choosing among options but as their only door still open – and building back toward them.
This exemplifies why the second pillar of the World EBHC Day 2026 Campaign matters – accountability (something Linda was denied) should be part of the story, not a separate concern. When a clinician misdiagnoses a patient, there is a record, a regulator, and a route to recourse. Yet, when a chatbot makes a misdiagnosis, there is a conversation log and a clause disclaiming medical advice. We have built a duty of care into every node of the formal system, and no such duty into the tool that growing numbers of people now consult first. For Linda, that absence of accountability was not abstract. It cost her four years of scans and antibiotics, treating a diagnosis that no one was answerable for.
None of this argues for keeping AI away from people like Linda. The computer or the phone is not the problem, and pretending otherwise insults the resourcefulness that led her to this solution. Instead, we must recognise that a free diagnostic tool reaching people where a clinic cannot does not mark progress. What people really need is AI assistance alongside care from a trained, accountable medical professional. Putting people at the centre means making the tool a route into the system, not a holding pen outside it: triage that ends in an appointment, advice that names its own limits, and AI that knows when to say ‘this needs a person’ and points to one within reach.
The World EBHC Day 2026 Campaign is right that the choices around the usage of AI tools are ours, but Linda did not have that choice. She got the only option available in her circumstances, and it cost her four years.
References
Data Reportal (2025). Global Digital Report for South Africa in 2025. https://datareportal.com/reports/digital-2025-south-africa
Human Sciences Research Council (2020). South African Social Attitudes Survey (SASAS). https://hsrc.ac.za/special-projects/sasas/
South African Government (2020). Report of the Presidential Commission on the Fourth Industrial Revolution. https://www.gov.za/sites/default/files/gcis_document/202010/43834gen591.pdf
To link to this article - DOI: https://doi.org/10.70253/RFDK8689
Conflict of interest
Nasreen is a member of the World EBHC Day Steering Committee.
Disclaimer
The views expressed in this World EBHC Day Blog, as well as any errors or omissions, are the sole responsibility of the author and do not represent the views of the World EBHC Day Steering Committee, Official Partners or Sponsors; nor does it imply endorsement by the aforementioned parties.
