4
AI chatbots could help safely diagnose patients ahead of GP visits, study finds (www.independent.co.uk)
submitted 3 days ago by to c/unitedkingdom@feddit.uk
5 comments fedilink hide all child comments
 

The study found that AI’s suggested diagnoses matched the eventual confirmed diagnosis in more than half of cases

So it got it wrong half the time?

However, one conversation was later rated “somewhat harmful” by a doctors after the chatbot listed a possible cancer diagnosis which caused the patient anxiety.

In another conversation, the chatbot hallucinated, and clinical information was added in five interactions.

This ain't it, chief.

you are viewing a single comment's thread
view the rest of the comments
[–] 12 points 3 days ago

Accuracy reached 75 per cent when considering the chatbot’s top three suggestions, while none of the conversations triggered safety stops.

I mean I know doctors make mistakes too, but 75% if you get 3 guesses seems quite lacking. Especially if you then need to confirm with a doctor anyway.

And yeah, the hallucination part is insane and surely unacceptable.

Not to mention I'd need to read the study but I wonder what the cases and symptoms were, and if doctors ended up agreeing with the top 3 due to having read the transcripts beforehand. And we also know even experts can be fooled into thinking someone else is an expert if they talk well (see: Dr Fox effect, and related studies).

  • permalink
  • fedilink
  • source
    •