Question to the Department of Health and Social Care:

To ask the Secretary of State for Health and Social Care, what steps he is taking to ensure that non-deterministic outputs from AI tools used in (a) diagnostics and (b) treatment planning can be (i) explained and (ii) validated.

In the United Kingdom, the majority of artificial intelligence (AI) products being used in health and social care are regulated as medical devices. This means they are subject to stringent safety, performance, and efficacy requirements, primarily set out in the Medical Devices Regulations 2002, with robust monitoring by the Medicines and Healthcare products Regulation Agency. These requirements also apply to non-deterministic AI systems that are medical devices used in the National Health Service, such as large language models.

The National Institute for Health and Care Excellence (NICE), an arms-length body of the Department, makes recommendations on the safety and cost-effectiveness of products for health and social care services in England. NICE routinely evaluates medical technologies, including innovative AI-enabled technologies, which are assessed in NICE’s Early Value Assessment programme, for those technologies that are most needed and in-demand.

NHS England is also developing guidance for those adopting or considering adopting AI technologies, which is expected to support the safe uptake of these technologies and alleviate concerns that clinicians currently have when buying and using these technologies.

Although many generative AI systems need minimal intervention from users, clinicians still need to review and approve the text which has been produced. AI tools are to be used to support clinicians with their own decision making, rather than to make decisions on the clinicians’ behalf. AI augments, rather than replaces, human expertise.

To ensure that clinicians understand the tools they are working with, we have developed tailored guidance for users and regulators. Through public dialogue, we’ve involved patients and the public in deciding how and why access to health data should be granted for AI research and development. In addition, to bridge the gap between policy and practice we have launched the AI Ambassador Network, which currently has 5,000 members.