How Customers Get Around AI’s Poor Math Skills

Healthcare providers and artificial intelligence vendors have touted AI’s potential to automatically interpret medical test results. But the technology still faces a major limitation: It struggles to do basic math. For example, Anthropic’s Claude AI chatbot often can’t answer correctly whether a certain number falls within a given range.

That’s a big problem for doctors using AI for medical tests. Without the ability to decide whether a number is within a range, it’s impossible for the AI to determine whether a biological metric like blood acidity falls within a normal interval, according to Dr. Praveen Meka, a physician at the Dana-Farber Cancer Institute in Boston.