AI-driven health policies are taking away the needy in one of the world’s poorest countries
Kenya has been promising wide access to affordable health care. But a new investigation has revealed that its algorithm-driven system is making life harder for the people it should be helping the most. According to The Guardian, Africa Uncensored, and Lighthouse Reports, the Social Health Authority of Kenya’s new system uses a predictive machine learning algorithm to estimate how much people should pay for public health insurance.
The program was first introduced in October 2024 as part of President William Ruto’s promise to increase access to healthcare for many of Kenya’s informal workers.
How the algorithm is hurting Kenyans
The problem is how the system calculates what people can afford. Kenya’s SHA program uses proxy means assessment, which is a method that estimates income based on household details such as roofing materials, toilets, livestock, family size, and other living conditions. The investigation revealed that the program overestimated the income of poor families, while underestimating wealthier citizens.
One SHA volunteer described visiting homes in Nairobi and watching people who were already struggling to find food receive premiums beyond their means. Some faced charges amounting to 10% to 20% of their minimum income, according to the report.
When debt prevents treatment
The consequences are real, and the situation seems dire. Kenyans without private insurance who cannot pay their SHA premiums are at risk of being turned away from health facilities or incurring steep hospital bills. The report cites reports of seriously ill people missing treatment because the program says they owe more than they could pay. One single mother said her monthly contribution was set at 3,500 Kenyan shillings, while others reported a huge jump from what they were previously paying under the old system. So the new policy is costing lives.
Although Ruto described the system as AI-powered, the report notes that it does not use ChatGPT-style generative AI. It uses predictive machine learning, built around a decades-old policy tool that has long been criticized for misrepresenting who qualifies for aid. Many have called the system flawed and unbalanced even before it is implemented.
More than 20 million people are registered with SHA, but only about 5 million pay their premiums regularly. Hospitals are also reporting huge shortages as reimbursements have not been paid. This is the danger of algorithmic welfare programs.
