The UK government isn’t underpromising about what AI can do in the NHS. The Ten Year Plan (TYP) is almost breathless in its enthusiasm. AI tools, it says, will improve diagnosis and treatment, accelerate clinical trials, optimise hospital workflows, facilitate 24-hour access, empower patients and enable personalised care plans. AI will also relieve staff shortages, reduce burnout and be a “trusted assistant” to doctors and nurses—and much, much more besides. And it will even save money—six pounds for every pound invested, the government claims.

The TYP also claims “the NHS is the best-placed system in the world” to take advantage of AI. Here, we need some salt: the conversations I’ve had in recent weeks with managers, tech professionals and clinicians paint a very different picture. It’s one where staff feel confused, unprepared and often disengaged, digital and data teams are overstretched and under-resourced, and many people with the skills and experience to work with this transformative technology are losing their jobs.

But don’t take my word for it. In December, the tech intelligence network Digital Health published a survey of NHS digital leaders. It should make uncomfortable reading for the AI enthusiasts in Victoria Street. Almost two-thirds said they felt unprepared to deliver the TYP’s digital commitments and a whopping 96% said workforce pressures were holding back progress. Only a third felt ready to deliver key government programmes like the Single Patient Record and the Federated Data Platform.

Leaders variously described their organisation’s digital readiness as “poor”, “chaotic”, “immature” or “patchy”. And preparing to implement AI was right at the bottom of their priority list.

AI doesn’t work straight out of the box. Which is why there’s such a big gap between the government’s promises and the reality on the ground. Developing, testing, deploying and learning to use AI systems in healthcare is a time consuming and labour intensive business.

The assumption that “you buy the technology and that’s the job done” is wrong, says Pritesh Mistry, research fellow in digital technologies at the King’s Fund and a former tech manager at Guy’s and St Thomas’ trust. “There’s a whole wrap-around of enabling factors that are needed to support the technology actually being used.”

As well as making sure new technology is compatible with existing systems, which hasn’t always been the case, he says, you need a deep understanding of the problems AI systems are supposed to solve. “We often hear about technology being implemented without considering the work people do and how they’ll have to change the way they work to get the best out of it,” Mistry explains.

“If there’s insufficient capacity—not enough time for people to do the training, to have exposure to the technology, to get the right support to use it, then it doesn’t get used,” he adds.

Experience shows that “if you just hire IT people to implement a solution, it doesn’t work,” Mistry continues. But involvement of non-tech staff in AI projects has been “patchy”, with development often left to IT services and a few “very digitally involved and experienced people”, he says. Ensuring technology is fit for purpose and usable by staff comes “from co-production, from working with the people who are going to be using the technology.”

That isn’t happening enough, says Tom Micklewright, a GP and digital transformation lead for Cheshire and Merseyside ICB. “It’s already hard enough to pull clinicians off the frontline, even for mandatory training, and this just becomes another thing to take them off”, he explains. “And if I’m going to get involved in making sure AI is safe, I can’t do that in a lunch break. It’s a lengthy bit of work.”

A study of the early use of AI in chest diagnostics, published last year by The Lancet, found procurement and deployment of AI took longer than expected, in large part due to high clinical workloads, limited AI knowledge and a lack of time for staff to participate. “AI may not address service challenges as straightforwardly as policymakers anticipate,” the study concluded.

Failing to engage clinicians and managers creates a “bottleneck” in the implementation of any new technology, warns Karl Grundy, an industry expert and CEO of Digital Health. “You need them involved in terms of software design and usability, but also for optimisation within the organisations and to win hearts and minds,” he says.

This “buy-in” is important, Micklewright adds, because “as a clinician, I’ve got to trust the tech if I’m going to use it”. Involvement also deepens understanding, helping to ensure the technology is used “responsibly”, he says, pointing to a recent example where GP trainees were found to be using ChatGPT to look up medical queries and summarise consultations.

Mistry has warned about the risk of a “disenfranchised culture” towards technology, where previous unmet promises and failure to involve staff in innovation breed scepticism and disengagement. Going by my conversations with non-tech managers about AI, this threat is real. A typical response was a snort, followed by a swift retort about computers that take ten minutes to start up and WiFi that doesn’t work through walls.

“Staff will be promised a step change that technology will unlock, but in reality often it doesn’t get realised,” Mistry says. “There’s money for the technology but not for the change—people are often unsupported for the change to happen.”

This is why, Grundy says, AI came bottom of the list of priorities in Digital Health’s survey of tech leaders. “We still have organisations going live with new EPR [electronic patient record] systems, or people that don’t have full meds management and things like that,” he explains. “If you haven’t got an EPR system… and you’re not collecting data on the health of your residents, there’s no point in having an AI system.”

In primary care, “contractual gets done first,” Micklewright explains, pointing to the momentum behind online patient records and prescription ordering—both enshrined in GPs’ contracts. “And second, it’s demands from NHS England”. But there’s been little direction from the NHS England on AI, he says. “It’s just been: ‘Start using AI. We’re not going to tell you which [systems], what to use it for or how to fund it—and we’re certainly not going to give you any money to try things out.’”

Another big threat to the government’s AI ambitions seems entirely self-inflicted. Redundancies triggered by the government’s NHS reforms are cutting a swathe through the service’s precious reserves of digital expertise just when they’re most needed. Some tech managers I’ve spoken to—mostly off-the-record for obvious reasons—seem close to despair.

A 2023 report by the Commons Health and Social Care Committee concluded there weren’t enough digital skills in the NHS to realise government ambitions. Action is needed, the committee said, both to boost the skills and confidence of NHS staff generally, and to expand the digital and data workforce.

This is another challenge for the new workforce plan. Mistry says that, while the NHS can’t compete on salaries, “there’s an opportunity” to grow the digital workforce by offering jobs that are “more satisfying”, have “more impact” and provide better work-life balance. “I think it’s probably needed, but we don’t know. That’s part of the problem,” he adds. “The NHS hasn’t done a good job of understanding how big the [digital] workforce is and what we’re going to need in the future.”

Some managers fear the NHS could get stuck in an outsourcing spiral as it tries to ramp up AI adoption in the years ahead. If digital staff “are spread too thin, we won’t be able to deliver anything, and that provides a further excuse to dismantle,” says Peckitt. This could leave the NHS totally reliant on big tech firms, without even enough in-house capacity to function as an intelligent customer.

The danger here is the NHS ends up “outsourcing its competence” as well as the project, Peckitt warns. NHS staff “need a level of maturity” to use AI tools, because they can often produce “plausible” results or “something that’s almost right”, he says. “But you need it to be absolutely right. So you need someone who knows what that is.”

The government’s widely-ridiculed AI Skills Hub—developed by consultancy firm PwC for the Department for Science, Innovation and Technology at a cost of £4.1 million—is a case in point, he says: “They inadvertently gave an excellent education on the dangers of artificial intelligence by outsourcing their thinking and having something generic but plausible-looking generated for them without inspection.”

Mistry says there is “qualitative evidence” that organisations like Alder Hey, Leeds Teaching Hospitals and Great Ormond Street, which have invested in their own digital workforce, technology skills for clinical staff and technology infrastructure, “are able to do more with technology and deploy it to have greater impact”.

And he cautions the government against moving too quickly with AI deals to avoid “cementing in a legacy technology”. AI is changing so rapidly, he says, that “if you’re locked into a five-year contract, how are will you benefit from technology that’s going to be so much more advanced in two years’ time?”

Despite the government’s lofty ambitions, nine months on from the Ten Year Plan, job cuts and the lack of a coherent national strategy for delivery have left digital teams and AI-enthusiastic managers and clinicians adrift in a sea of uncertainty.

Peckitt points to the success of the London Secure Data Environment, a project linking up the health records of Londoners, as an example of what NHS digital teams can deliver. “We can do remarkable things. But to get rid of people at this time and leave people without a clear direction during a time of technological inflection is irresponsible,” he says. //