I am an out-of-hours GP, and I am glad to see a news report on this issue, which has been causing me increasing concern at work. Doctors’ AI scribes get names of drugs and diagnoses wrong, NHS watchdog warns (31 August). There are a number of problems with AI, particularly its tendency to misunderstand and contradict itself, which can lead to dangers for the patient.
I will concentrate on the supposed ‘time-saving’ aspect. Many patients I see have already been assessed by someone else on the phone, and I am used to reading through a triage assessment before consulting them. My heart sinks when I see that AI has been used—consultations are invariably long, with duplications, and sometimes contradictions.
Often, any history I take turns out to be significantly different from what is listed. I can never trust that it is correct—I find meaningful errors far more often in AI consultations than in those typed out by colleagues. You may blame the original clinician, but anyone who has tried to proofread something under time pressure knows how difficult it is.
So how can this save time when the first clinician has to check through a long, repetitive account, and the second clinician can’t trust what is written and has to check it again? Turning a complex patient history into a clear account for colleagues is a difficult skill to learn, as medical students understand. If we do not maintain this skill, we will lose it—and so will our patients.
Dr Mary Gibbs, Manchester, shares first-hand experience of the risks of AI diagnosis and medication errors. A consultant she worked with as a secretary tried voice recognition software for a week. It failed when ‘I recommend two weeks on lansoprazole’ was misinterpreted as suggesting a holiday in Lanzarote.
Debbie Cameron Formby, Merseyside, highlights similar issues.
Source: The Guardian
