Last Thursday I sat with a physician-scientist from Stanford, a biomedical engineer from Uganda, and a perioperative nurse from Lagos. We talked for 90 minutes about AI and healthcare in Africa. The conversation was supposed to be structured. It became honest instead.
I want to share what I took away.
The calibration problem
Faith Emoruwa works in robotic surgery in Lagos. She told us about a dialysis center that had to abandon its EMR system because the software kept flagging their patients as critically ill. The reason was simple. The patients’ PCV levels were running at 20 percent, which is the accepted clinical standard at that center for dialysis patients, but not normal by the parameters the system was trained on. Every time the clinical team tried to proceed with patient care, the system stopped them. A big red warning sign. A Western system making clinical judgment calls in a Nigerian context it had never been designed for.
They eventually switched to an EMR built by a Nigerian company. The problem went away.
This story matters beyond that dialysis center. Only about 2 percent of the data currently feeding AI models comes from Africa. The parameters these tools use to judge normal versus abnormal, healthy versus sick, were built predominantly on Western populations, Western clinical contexts, and Western infrastructure assumptions. When you deploy those tools in Lagos or Kampala or Nairobi, you are not just dealing with a technology gap. You are dealing with a calibration problem.
AI does not know that a significant portion of West Africans have lower white blood cell counts that would be flagged as dangerously low on Western charts. It does not know that a dialysis patient in Lagos operates within a clinical range their counterpart in California would treat as an emergency. It does not know what the generator backup situation looks like, or what it means when the robot overheats and the technical team is 8,000 miles away.
The tool is not wrong. It was just never designed for this.
What cannot be outsourced
Dr. Chidi Akusobi made a point that has stayed with me. He said that in the age of AI, two things must never be outsourced from medicine: critical thinking and empathy.
AI can retrieve. AI can synthesize. AI can document, flag, scan, and summarize. But the ability to look at what a tool generates and say this feels right or something is wrong here is not a skill the tool can give you. It is a skill built through the hard labor of learning, of wrestling with problems, of being wrong and correcting, and of being wrong again.
And when a patient enters an operating theater at 3 in the morning, terrified, about to have a needle inserted into their spine, they are not comforted by a machine. They need the human being at their side to say: I see you. You are safe. We have done this before. That is not inefficiency. That is medicine.
I told a story during our conversation about a hotel in London that replaced its doorman with an automatic door. Months later they noticed a drop in bookings. When they investigated, they realized the doorman was not just opening a door. He was the face that smiled as guests arrived. He was the person who ran out with an umbrella when it was raining. He was doing things his job description had never captured. We stripped out the human and called it progress. Then we counted what we lost.
Infrastructure before sophistication
Dr. Maureen Etuket told us about a hospital managing director in Uganda who dramatically reduced maternal mortality. Not by introducing new devices. Not by deploying AI. By improving training.
Just training.
The gap between a pilot and a patient is not always a technology gap. Sometimes it is a training gap. Sometimes it is a calibration gap. Sometimes it is the gap that opened when USAID funding was pulled and nobody had a plan for what comes after.
Before we ask how to introduce AI into African healthcare, we need to ask whether the system it is entering is ready to absorb it. A powerful tool deployed into a broken system does not fix the system. It adds complexity to it.
The work ahead
By 2030, Africa will have released close to 375 million young people into the labor market. Most of them are being trained for a world that no longer exists. The recall-based education model was designed for a time when information was scarce. That time is over.
We need to train healthcare workers to think critically, not just recall. To adapt, not just execute. To understand the edge cases and failure modes of the technologies in their hands, because the technical team is not always going to be available at 3 in the morning.
We need to feed African data into these models, or build our own. We need policies that make healthcare investment attractive to the private capital that is currently going elsewhere. And we need conversations like the one we had last Thursday. Not conferences. Not keynotes. Real conversations, led by people doing this work on the ground, drawing from the lived reality of actual African clinical settings.
AI handles what can be specified. Humans handle what has to be felt.
The continent has more to gain from AI in healthcare than anywhere else in the world. But potential does not transform itself into reality. People do.
