New research suggests that medical AI chatbots are woefully unreliable at understanding how people actually communicate their health problems.
As detailed in yet-to-be-peer-reviewed study presented last month by MIT researchers, an AI chatbot is more likely to advise a patient not to seek medical care if their messages contained typos. The errors AI is susceptible to can be as seemingly inconsequential as an extra space between words, or if the patient used slang or colorful language. And strikingly, women are disproportionately affected by this, being wrongly told not to see a doctor at a higher rate than men.