Britain’s industry regulator has called for updated legislation to govern the use of artificial intelligence in the National Health Service and other medical settings. The Medicines and Healthcare Products Regulatory Agency (MHRA) recently released 44 recommendations designed to modernize its policies as AI integration into healthcare accelerates.
Lawrence Tallon, chief of the MHRA, told the BBC that AI will soon become a routine component of NHS services. He emphasized that adoption must proceed in a manner that preserves patient trust and confidence.
The report was developed by an independent commission and incorporated feedback from over 12,000 individuals, including clinicians and patients. Key proposals include continuous monitoring of AI products with removal from approval if they malfunction or lose effectiveness. The recommendations also grant patients the right to know when AI is involved in their care and provide the authority to penalize developers whose products fail to meet standards. Additionally, a new “L plate” system would allow healthcare professionals to trial new AI models under strict supervision.
Tallon noted that the current regulatory framework was established during an era focused on physical devices such as hip replacements and stethoscopes. While existing rules may suffice for simple AI tools designed to identify known symptoms on scans, they are inadequate for more complex models. Unlike traditional medical products, AI systems continue to change after authorization, learning and adapting as new data is fed into them.
The challenge of regulating AI is global, Tallon acknowledged, stating that no single country currently possesses a perfect regulatory framework.
AI-powered note-taking tools, often referred to as scribes, are reportedly utilized by 40% of general practitioners in the UK to record consultations and generate reports. However, research from the University of Edinburgh suggests that patients may be less inclined to disclose sensitive personal information, such as substance abuse history, if they are aware that AI is processing their conversations.
Professor Henrietta Hughes, a GP who served on the report commission, reported that while many patients are comfortable with AI in consultations, some choose to opt out. “Some say, ‘I don’t want to talk to a robot,’ and that is also fine,” she said. She added that while these scribes can make errors, it remains the doctor’s responsibility to correct any inaccuracies.
Currently, AI is predominantly used to assist with administrative duties or support symptom identification under human oversight. Nevertheless, there is significant enthusiasm regarding its potential in health and medical research. Professor Alastair Denniston, an ophthalmologist involved in the report, described the technology as an exceptional opportunity comparable to historical breakthroughs like antibiotics and MRI scans.
Despite this optimism, concerns persist regarding potential biases in patient data used for training models and the risk of AI chatbots providing incorrect medical advice.
I worry about bias in training data. Will these new rules actually force developers to prove their models are fair for everyone?
Continuous monitoring is a smart idea. Static approvals for self-learning algorithms just don’t make sense anymore.