A panel of health system managers, clinicians and policymakers discussed the pain points surrounding artificial intelligence (AI) in healthcare during a dedicated event in Basel this week.

Speaking at the Intelligent Health conference on 11 September, representatives discussed the potential benefits of AI in healthcare and the challenges they faced along the way.

Dr Stephanie Kuku, senior research fellow at University College London Hospitals, argued that companies with proposed machine learning solutions should be challenged to prove they could make a difference on the frontline.

She added that a lack of infrastructure meant that, in many instances, AI tools failed to make a valuable impact on patient outcomes.

“You’ve got to have the infrastructure to make sure patients get the care they need,” said Dr Kuku.

“There’s no point having a tool that tells you that you need to go to hospital if you can’t get to one.

“There are a lot of examples, sadly, of tools that are not making a difference to patient outcomes”.

On the issue of AI bias, Dr Kuku argued that AI algorithms should be trained using data representative of the demographics they would ultimately serve.

“If you’re going to sell algorithms to NHS, you need to make sure data is representative,” she said.

“The more we can generalise these algorithms, the more we can think about how it can help as many people as possible.

“Every team where you built models needs to be intensely multi-disciplinary. The more multi-disciplined your team is, the better, safer and more successful you’ll be in terms of both patient and financial outcomes.”

Dr Kuku’s sentiments were echoed by Charles Alessi, chief clinical officer at HIMSS International, who said deployment processes needed to be “finessed”.

Alessi argued that potentially beneficial new technologies were being killed off before they could have an impact due to of botched deployments that “completely destroyed business models”.

“It fills me with enormous sadness,” said Alessi.

“I’ve seen wonderful technology introduced into healthcare systems all over the world, followed soon after by the death of these wonderful new technologies, because people don’t think about how to deploy them.

“We need to think: how does it influence a business model, or a doctor’s consultation? You remove one brick and suddenly the whole thing falls apart. What you tend to get is unexpected consequences, with boring regularity.”

Acquiring the right talent to deal with the vast amounts of data involved in AI remains a pervasive issue.

Karl Goossens, associate partner at data analytics firm QuantumBlack, said presenting data insights in easy-to-digest forms was key to demonstrating the benefits of machine learning to both healthcare professionals and the public.

“You might have a great data model and great output, but a lot of people aren’t analytic-savvy,” explained Goossens.

“One aspect is convincing them [your data] is valid; the second is thinking about how you present the information to the end user and trying to make something that is quite complex, understandable and intuitive.

“If you think about consumer devices, they are self-explaining. At the end of the day, we want to get to something like this.”

Delegates at the two-day conference also heard from Microsoft’s corporate vice president of healthcare, Dr Peter Lee, who said “startling advances” in natural language processing technology promised to ease the overwhelming burden of paperwork faced by doctors.