Hazel Tang A science writer with data background and an interest in the current affair, culture, and arts; a no-med from an (almost) all-med family. Follow on Twitter.
The AIMed’s deep dive Breakfast Briefing in July 2019 examined how Switzerland’s largest healthcare organisation – University Hospital of Bern (part of the Insel Group) – is making advances in the applications of AI in medicine.
The panelists’ technical, legal, and clinical expertise shed light on the current issues and future challenges facing those leading the transformation of healthcare through AI.
The discussion started with the panelists’ reflections on what place AI should have
in the healthcare industry and what the teething problems associated with integrating it into classical practices are.
Prof. Dr Roland Wiest touched on image analysis when looking for inflammatory brain disorders to underline the point that humans make errors (sometimes as much as 25% of the time). AI is good for standardisation. But as we increasingly turn to technology, there is a risk of us becoming too dependent on AI. The panelists considered the critical issue of what will happen if we become over reliant on AI.
Allocating accountability to a human for the failure of AI remains a contentious issue according to Dr Ursula Widmer, a legal expert and founding member of the law firm Widmer & Partners, because it is unclear which human should be held accountable. The current process leaves it ambiguous as to whether the provider, the engineers, the doctors using the technology or a whole host of other actors are responsible for the technology’s failure. It is a mistake to think of AI as ‘perfect’. The maths and algorithms behind it allow the machine to make a ‘best guess’ as noted by Dr Christian Westermann and as such we should be cautious in treating it as a momentous breakthrough set to replace clinicians altogether. Instead, the goal should be to make the best use of the skills both humans and AI have to offer.
From a commercial perspective, innovation needs to follow a set of ethical principles before technology is implemented. Prof. Dr Wiest’s first consideration as a clinician is how to meet guidelines – there needs to be procedures on how to use technology and handle them in case they fail. He argues that this thinking needs to be brought into society as is being done by the European Academy of Neurology. The essential thing is to show that using AI is something people can trust. Quality control and better investment in security were identified as steps towards Responsible AI.
‘Responsible AI: A Global Policy Framework’ is an in-depth review of eight policy principles related to ethical guidelines encouraging the responsible use of AI. These include accountability, the transparency of AI algorithms, and non-discrimination. The technical perspective of Responsible AI, outlined by Dr Westermann, is similar and considers the fairness of algorithms, the interpretability of outcomes, and the robustness of the technology.
On the topic of using AI responsibly, the panel covered issues with the current educational system in preparing clinicians sufficiently. The Certificate of Advanced Studies attempts to teach students the basics of programming to help them further understand AI models and their applications. It is the first of its kind. However, Prof Dr Wiest maintained that the legal consequences, ethical principles, and technical aspects of using AI need to formally be brought into the curriculum.
For Dr Widmer, privacy is also, by design, necessary in AI education. This is something she claimed hospitals do not invest enough money into in Switzerland. And this is
why the panel warned against the threat of adversarial attacks – manipulating the data to make an algorithm reach different conclusions. Better staffing and investment in security could protect against this.
The panellists gave insight into why they think Switzerland should be seen as an AI haven. They agreed it has the expertise, the infrastructure, and the trust to deliver AI responsibly. The density of the country brings technical and clinical thinkers close together, meaning Switzerland is in a privileged position to provide cutting-edge technology within a framework of guidelines.
THE PANELISTS
Moderator
Dr Fried-Michael Dahlweid, Chief Technology & Innovation Officer, Inselspital, University Hospital Bern
Presenters
Prof. Dr Roland Wiest, Professor of Advanced Neuro- Imaging, Inselspital University Hospital Bern
Dr Ursula Widmer, Partner, Widmer & Partners
Dr Christian B. Westermann, Leader Data & Analytics, Partner, PwC Switzerland
