DeepMind has begun a public consultation with patients and healthcare workers to receive feedback on plans to introduce its artificial intelligence technology to the National Health Service (NHS).
DeepMind – a London–based startup acquired by Google in 2014 - specialises in the creation of deep learning algorithms, and was responsible for the the AlphaGo programme which beat Go world champion Lee Sedol earlier this year.
It has also turned its attention to applying its technology to help modernise the ailing health service by digitising processes and making better use of data.
"This is a new space for us, and we have set out to involve as many people as possible in our design process from the outset," said Mustafa Suleyman, DeepMind co-founder, at a public forum event at the company's London offices on Tuesday.
DeepMind Health's initial projects have included a partnership with Moorfields Eye Hospital to apply its algorithms to the diagnosis of common eye diseases, and with the Royal Free Hospital to develop an app that improves the detection of acute kidney injuries.
However, while its initial endeavours have attracted interest against a backdrop of an under-funded NHS, it has also drawn criticisms over plans to use confidential patient data that is necessary to train its algorithms. This has included an investigation by the Information Commissioner's Officer around data encryption procedures following a public complaint about the company being given access to 1.6 million patient records at the Royal Free.
Addressing audience comments, Suleyman promised that one of its core philosophies would be ‘do no harm’ as it seeks to allay concerns around data governance, privacy and the corporate interests of parent company Alphabet.
“This has to be a mantra that we repeat and becomes an inherent part of our process and should be the first measure of success before any deployment, or before we attempt to demonstrate any utility and patient benefit,” he said.
Suleyman added that openness and transparency would be key to winning over those who are sceptical of DeepMind’s plans. He promised to continue to publish the results of its work, both its algorithms and “clinical peer-reviewed evaluations of the measurable impact we have had”.
While the specifics of its long term commercial aims and business model were less clear, Suleyman indicated this would be results-based, reflecting the success of projects.
“Of course that is important, we have to build a sustainable business model around this," he said. "We have been clear about this from the outset.
“What we have avoided doing is nailing that down too early and luckily we have the resources to explore what the most effective business model would be around our intervention, so that essentially we get paid when we deliver value."
He added: “What we would like to do is for at least some proportion of what we get paid to be connected to the actual concrete outcomes. That is another way in which patients can play an active role in helping us to identify what the important metrics are to patients, as well as the hospital in terms of its efficiency. That is going to be a really challenging aspect of the model.”
DeepMind is not offering a silver bullet for the problems of the NHS however. Suleyman acknowledged that its projects remain at very early stages.
One example cited was a patient portal smartphone app which would allow those receiving treatment to access appointment information and other data as well as potentially interacting with clinical staff remotely. It is an exciting prospect, yet Suleyman stressed that this is just a concept at this stage, and is “months if not years away” from actually being built.
He also recognises that ageing hospital systems will present a bottleneck to development, with DeepMind's technology still reliant on integration with legacy backend software.
He said: “That is a challenge for us because we obviously have to interact with those systems and even if information moves into our ecosystem and we can be absolutely confident that this is entirely secure and transparent and auditable and verifiable and so on, we are still interfacing with a system that is many years behind the cutting-edge with respect to information security processing.”
He added that a lot of the technical systems that store and process data were “genuinely developed for completely other sectors, not even healthcare".
"It is pretty remarkable when you look at it,” he said. “There have been clinical systems today used to run entire hospitals which at their heart were originally designed for logistics systems or accounting systems or for manufacturing and have been adapted first to a US context and subsequently to the UK and NHS.
“That means we are dealing with quite old and quite brittle technologies that really haven’t been subjected to improvement for a long time.”
According to Suleyman, reliance on ageing infrastructure has throttled previous attempts to create digital services in healthcare.
“Frankly speaking that is why there hasn't been a great deal of innovation in the sector. There are all sorts of startups who have incredible ideas and would love to deploy these sorts of solutions and I think the main obstacle they have been facing is the barriers to entry that some of the existing larger players have put around access to data."
DeepMind will not be immune to these challenges either, though Suleyman is confident that industry collaboration will help.
“We are going to face those issues too but I think we are maybe in a slightly stronger position to try and develop a more open innovation ecosystem around this data, obviously controlled by the data controller, the trust and by the patient.”
Clearly DeepMind is under no illusion of the scale of the challenge it faces, both from a technical point of view and in terms of wider public and regulatory perspectives. But there is no doubt that its technology offers intriguing possibilities for improving healthcare in the UK.
The next steps for DeepMind Health will be to continue consulting with staff and patients as often as possible, with a patient involvement strategy document to be published in January 2017.
"This is really just the beginning," Suleyman concluded. "There will be many, many more forums where we can take your critical feedback and show you how we have improved and updated in the months in between these kinds of sessions."