The successful use of AI in rare disease identification is challenging. Good outcomes still depend as much on human intelligence as AI. Two potential points of jeopardy, while distinct, have their roots in a common theme: not measuring success correctly.

The first we’ll refer to as “When 95% accuracy is 100% Wrong”, and the second we’ll refer to as “Rare Disease Identification is not Diagnosis.” We describe what’s at stake in each case and outline a plan for transforming these two high risk points into high leverage points in your next AI-driven rare disease project.