TL;DR

A family from Bath say ChatGPT helped them reach a diagnosis for their 19-month-old daughter, Lily. In January, tests confirmed MSMDS (Multisystemic Smooth Muscle Dysfunction Syndrome), a genetic disorder the BBC says only six people in Britain are known to have. Genetic Alliance UK welcomes AI’s promise but warns that rare conditions are where chatbots are least reliable.

From symptoms to a name

Lily had heart surgery at four months old for a condition nobody could yet identify. Her mother, Rosie, typed her symptoms into ChatGPT, and it returned MSMDS. When she raised it with doctors, she said, “some of them had never heard of it; others said it is so rare it’s very unlikely”. Tests later confirmed it.

A specific variant of the ACTA2 gene causes MSMDS, which can damage organs including the heart and kidneys. Lily’s pupils are permanently dilated, so she wears sunglasses outside. She is one of about 70 known cases worldwide, and doctors at Great Ormond Street and at Bristol’s children’s hospital oversee her care. Her parents founded ACTA2 Alliance UK, a charity for the condition, and are raising £30,000 towards early work on a less invasive gene therapy.

Why experts urge caution

Nick Meade, chief executive of Genetic Alliance UK, stressed that the case was unusual and that the charity knows of few diagnoses reached through AI. His concern is structural. “Machine learning models learn mainly from whatever appears most often in their training data,” he said, adding that the flaw is well documented and “rare conditions are, by definition, the kind of case it affects most”.

Dr Rebecca Payne, a GP who co-wrote an Oxford study on the subject this year, said in earlier BBC coverage that “AI just isn’t ready to take on the role of the physician”. She warned that asking a chatbot about symptoms “can be dangerous”, with the risk of wrong diagnoses and missed emergencies.

Why it matters

Lily’s case is unusual, but turning to chatbots for health questions is not. A King’s College London survey reported in August found one in seven people had used AI for health advice instead of contacting a GP. For a family without answers, a chatbot’s suggestion can be what prompts a test. The same tool can also send others down the wrong path with confidence.

Looking forward

The BBC contacted NHS England and both hospitals for comment. The harder question for the NHS is how clinicians should respond when a parent arrives with a chatbot’s suggestion: dismissing it outright may miss cases like Lily’s, and accepting it uncritically carries the risks Meade and Payne describe.