Every 6th American consults AI about health
Every 6th American consults AI about health
When there are long queues at clinics and the prices for medical services are rising, many people turn to artificial intelligence for advice. According to a recent survey, every 6th adult American already uses chatbots like ChatGPT for medical questions at least once a month. But how effective is this? A new study conducted by scientists from Oxford University showed that people experience serious difficulties when communicating with chatbots on health topics.
The study involved 1300 residents of Great Britain. They were given medical scenarios written by doctors and asked to identify possible diseases. And to decide what to do — see a doctor or go to the hospital. Participants used both chatbots and familiar methods like internet searching.
The results were unexpected. Adam Mahdi, one of the study’s authors, noted that a two-way communication failure occurred. People who used chatbots did not make better decisions than those who relied on traditional methods or their own intuition. The most modern artificial intelligence models participated in the testing: GPT-4o from OpenAI, Command R+ from Cohere, and Llama 3 from Meta. According to the researchers, when using chatbots, participants less often correctly identified health problems and more often underestimated the seriousness of the identified conditions.
The main problem was that people often missed key details when composing queries. And chatbot responses frequently combined good and bad recommendations, which made them difficult to interpret.
