AI – doing more with less in the NHS

From stock market jitters caused by the artificial intelligence (AI) model DeepSeek to Government announcements on the AI Opportunities Action Plan, it can seem that news about AI is everywhere. Research and healthcare are not alone in grappling with how we can make the most of this technology in today’s data heavy world. At this year’s Festival of Genomics, discussion of AI and the impact that it could have on genomics research and clinical practice was peppered throughout the packed programme. 

At an open roundtable discussion hosted by the British Society for Genetic Medicine, all attendees took part in smaller group discussions and shared thoughts to the whole group on questions posed by the organisers, a valuable temperature check on what the wider genomics community is thinking about AI. 

Using AI on a day to day basis

A show of hands from the attendees found that the majority had used AI in some capacity in their day to day work, from kickstarting software development and validation of software models, to research uses such as hypothesis generation in research and extraction of data from medical records. We all acknowledged that we’re using AI on a day to day basis, even unintentionally, with Google search AI summaries being a key example. 

Where could AI be most helpful in genomics? Variant interpretation and support for clinical decision making are areas where most work is being done, especially on single nucleotide polymorphisms (single letter changes to the genetic code) and whether they are contributing to the condition being investigated. However, for these uses to be most effective, more work is needed on how AI can support the analysis of different types of variants and their contribution to disease, such as structural variants, and that more diverse genetic data, which gives a more global view of human genetic diversity, is vital. 

Elsewhere at the conference, conversations focused on using AI to explore patterns in medical data and support diagnoses in conditions that are otherwise being missed – familial hypocholesterolemia, an inherited condition that causes high cholesterol levels, being one example. The use of AI to support variant classification and interpretation in rare diseases is another use that was discussed more widely, providing clinical scientists with support in analysing complex cases where the genetic cause is less clear. 

The winding road to implementation

Back at the roundtable discussion, many were keen on the value of AI in supporting more administrative tasks, for example information extraction from the literature, data mining and writing letters. However, thoughts on the ethics of AI and the potential impact on future generations in terms of data use including bias within models, and our level of tolerance for errors, were never far from the surface. How we navigate the varying international regulations around AI, with different jurisdictions taking a range of approaches, is just one area that needs more attention. 

What is inescapable is the need for robust infrastructure and computing power – needs we highlighted in our 2020 report Artificial intelligence for genomic medicine –  to underpin delivery of AI in research and clinical practice – not just to store, share and analyse data, but also platforms that will allow day-to-day users to interact with these powerful tools. Common frameworks and rules for benchmarking are a key component of this, validating AI to ensure that it is used in a consistent and safe way. 

The general consensus at the roundtable was that implementation of new technology into the NHS is challenging, and AI will be no exception. Anyone leading implementation efforts must continue to do the fundamental things well: make a clear plan and communicate it well to build trust within the workforce. AI can be really helpful for decision support, but people didn’t want to lose the human touch, and have AI replace decision makers. I sensed that the overall tone was one of cautious optimism, to keep learning, keep talking, and progress with care. 

The right skills get the most out of AI

What struck me about many of the uses of AI by researchers and clinical scientists, in addition to its use by software developers, was the link to a point made at the roundtable: to get the most out of AI, you need to know how to ‘drive’ it; you wouldn’t put a learner driver behind the wheel of a Ferrari. The greatest power and insights will come from those who are already very skilled in a particular area and who can not only use AI to turbo-charge their work, saving time when carrying out more mundane tasks, but also understand and work with the answers it gives. 

Ultimately, the take-home messages for me were that the NHS could use AI to do more with less, and that we shouldn’t underestimate the value it could bring to managing ‘boring but important’ tasks, freeing up clinicians’ time to focus on other areas. Regardless of the use, the need for skilled operators is vital, as is education and engagement to support staff to use these tools. I hope that the tone of caution, care, optimism and enthusiasm seen at the Festival sets the tone for the implementation of AI into genomics practice for the benefit of patients, scientists and clinicians.