AI developers cull electronic health records because they hold huge amounts of detailed data, Cho said. But those developers often aren’t aware that they’re building atop a deeply broken system. Electronic health records were developed for billing, not patient care, and are filled with mistakes or missing data. A KHN investigation published in March found sometimes life-threatening errors in patients’ medication lists, lab tests and allergies.