UK AI medical transcription is dangerous and doesn’t work
The National Health Service in the UK is trying super hard to get doctors to use AI voice recognition everywhere a human might previously have been. For transcribing patient appointments, they just sit a microphone down in the echoey consultation room, and it feeds the ambient sound to a chatbot.…
The National Health Service in the UK is trying super hard to get doctors to use AI voice recognition everywhere a human might previously have been. For transcribing patient appointments, they just sit a microphone down in the echoey consultation room, and it feeds the ambient sound to a chatbot. This then generates a version of what was said, in various formats. The NHS says: [NHS] Ambient scribing products, when combined with generative AI, can convert speech directly into structured medical documentation, such as notes and letters. What could go wrong? Official NHS watchdog Healthwatch England spoke to patients who’d gone so far as to check their AI appointment transcript. Surprise! These things make dangerous and stupid errors that humans wouldn’t. [Healthwatch; Guardian] One doctor said how: [BJGP Life] the scribe was adamant I told the patient they are to ‘continue their Prozac’ that I had prescribed, a medication neither I or any of my UK-based colleagues would likely have ever authorised and that had not been mentioned in the consultation or featured in the scribe-formulated transcript. Why are doctors even using these things? Mostly, it’s massive NHS pressure. The doctors are flat out just keeping up, and it’s cheaper than hiring people. Doctors can see more patients, faster! Quality is not a metric here. Checking the transcript for more than a second never happens, because the excuse for the AI is the doctors don’t have the time in the first place. In August 2025, the Medicines and Healthcare products Regulatory Agency (the MHRA) recommended doctors report AI scribe errors as medical device issues to the MHRA. [BMJ] But in June 2026, the MHRA ruled that transcription is just admin support, so lying chatbot scribes are not medical devices. Now, you might suspect political pressure in the months between. [MHRA] The Royal College of Physicians is concerned, and it wants stronger regulation on these things. [RCP] This is if the AI transcriber even works. It’s great if you talk like you’re on BBC London. But patients in Yorkshire are stuck talking to a robot if they want to make an appointment, and it literally doesn’t understand their accent. I’m sure that reduces patient load nicely. [Guardian] The vendors always claim that patients trying to get an appointment out of the AI can totally speak to a human any time. This turns out not to be true. One frustrated Pivot to AI regular tells us about his experience being forced to make an appointment through this sort of AI system. He’s a retired medical professional, too. He called his doctor with a severe belly ache. The Rapidhealth system deemed it not serious enough for a consultation. It turned out to be a perforated bowel leading to sepsis. He tells us the story: My consultant and surgeon as you can imagine was very pissed off. He said. “If we had seen you within a few days we would have done a quick keyhole surgery and you would be home in two days”. Instead I was bed blocking for 30 days under the constant care of real life savers. The AI in medicine is actively dangerous. But it sure saves on staffing! If you can tell your doctor “no”, tell them “no”. I certainly do. Yes, they’re trying this in NHS Scotland too. Video — PodcastSource: Pivot to AI — Published — Category: Business