How to humanise work with AI

Artificial Intelligence (hereinafter: AI) is on the rise in healthcare and is therefore an important focus area for Deloitte. In the future of healthcare, AI not only has consequences for the patient, but also for healthcare workers and their work. To prevent the focus on technology, Deloitte focuses on 'humanAIsing work', in other words: making work more human, with AI as an accelerator. However, AI is still a subject that healthcare needs to get used to and learn to deal with in a careful and practical way. And in view of the rapid developments in the field of AI - and also in the field of legislation and regulations about it - it is important that it picks up speed. Erasmus MC has given direction to this by means of a competence centre in order to be able to centrally steer the development in this area.

AI is an important topic for healthcare for several reasons, says Marly Kiewik, partner at Deloitte. "It plays a role in creating better access to care, in providing better support for healthcare workers and in controlling healthcare costs. We see three main areas of application for AI: in the primary care process, in the more indirect - including administrative - processes and for increasing patient engagement."

The first area of application is really about healthcare itself, about predicting diseases, for example or linking information from patient records, images and information collected by the patient - such as wearables and sensors - into healthcare processes.

The second is the back-end: the administration and logistics of healthcare. Examples include smart scheduling and scheduling based on patient requests and staff availability. But also centralising communication, which is still often done with various apps and devices.

Kiewik: "The third, patient involvement, can be considered, for example, of increasing the patient's control through self-diagnostics or being informed by a digital human or chatbot in addition to an employee. In all three of these areas, we see that experience is already being gained with AI. And it will mean a lot more for healthcare in the short term."

Barrier broken down

Also, Simon Vermeer, Chief Information Officer at Erasmus MC, observes that developments in this field are progressing rapidly. "The development of large language models has made a significant contribution to this," he says. "They break down the barrier that has long existed between humans and technology. We have found technology to be challenging for a long time, but thanks to applications based on such models, we can now communicate with computers and smartphones in our own language and also receive comprehensible responses. This is particularly what the healthcare sector has been waiting for, as many people working in this field are less familiar with technology."

In terms of medical content, a lot was already happening with AI in healthcare before those large language models were introduced, Vermeer outlines. For example, Erasmus MC is currently in the tendering process for a new alarm system. In addition, the application of technology to combat alarm fatigue among employees is very important for the hospital."

When we started with that process, we measured and signalled everything. That becomes unworkable for the employees. So we are now looking at the possibilities of AI to do this in a more targeted way. This must be done correctly, in accordance with laws and regulations. We are now gaining experience with this. Of course, that slows things down, but everyone understands that you have to be careful about this. An example of this is that we already use algorithms to predict no-shows at the outpatient clinics.

This is subject to less stringent requirements, so it is a good way to learn how we can apply AI at the front end, for the direct care process."

Reliable and explainable

A point of attention in AI development is that in machine learning it is not always clear how the computer arrives at an answer. "It shouldn't be due to bias or an error in the algorithm that leads to the wrong outcome," says Vermeer. Kiewik adds: "You have to be sure that you have reliable data and - where possible - explainable algorithms, otherwise you are completely dependant on the outcome that the computer gives."

Where 'explainable' will always have a limit, says Vermeer. "We really don't understand ChatGPT's algorithm," he clarifies. "That's what the discussion is about. We are gaining experience with AI by, for example, shaping the input/output equation in such a way that we know that we can achieve reliable outcomes with AI. Until then, we will not use ChatGPT for medical applications. But we do realise that we need to gain knowledge about it because it will certainly have a place in it. So we set up a platform to develop, learn and validate. There's a lot to consider."

Proceed with care

Vermeer says he fully supports the caution with which healthcare is proceeding. "But I also see that the use of AI in healthcare is irreversible," he says. "So we have to learn how to apply it responsibly and I'm sure we're going to be able to do that."

This will mean a big change for the employees, says Kiewik. "It's too easy to say that AI is going to take over routine work and thus provide more space for the interesting parts of the work," she says. "However, it's not just about routine tasks. It also means that healthcare workers and IT colleagues will become increasingly close to each other. The IT staff are more prominent, towards the primary care process. The relationship between the two intensifies.'"

Democratisation of IT

This process has been called the democratisation of IT, says Vermeer. "ChatGPT is a great example of that, it's a powerful
tool that is very easy to apply. But that's not the case for all AI applications. That is why we say: there are many vacancies in healthcare, for some of them look at people who know about them. If more employees become familiar with it, I don't rule out the possibility that the number of IT professionals will decrease and that employees in the primary process will be given a different profile."

This automatically has an impact on study programmes, Kiewik thinks. "And on the way healthcare hires people. In doing so, it will be important to look more at skills than at diplomas." Still, that will take time, Vermeer expects. "The methodology around job descriptions appears to be quite cumbersome. It turns out to be difficult to flip the switch in the mind that we need different competencies from healthcare professionals. However, with the increasing accessibility of IT applications, this is becoming feasible, especially for the younger generation.

Safeguarding

Erasmus MC has set up a competence centre to centrally steer the upscaling from pilot to structural embedding of AI applications. "This is necessary, because there are also a lot of laws and regulations involved," says Vermeer. "Also keep in mind that AI becomes a medical device the moment you start applying it in healthcare. Along this line, the Health and Youth Care Inspectorate, among others, will also monitor what the healthcare sector does with it. You really have to embed it in the organisation so that it becomes a routine part of business operations. The rapid developments in the field of AI are putting pressure on us to take rapid steps in this direction."

That is why coalitions of healthcare providers are also emerging to share knowledge and costs together", Kiewik indicates. And that is necessary, Vermeer believes. "Erasmus MC is an organisation with a large scale, but for small healthcare providers it will be a real challenge."

The European Union has now published the EU AI Act1. "It's a good development," says Vermeer, "because it provides frameworks. The act defines what constitutes the responsible application of AI, what it may and may not be used for, and what requirements must be set in this regard. This gives direction to the development that we are going through in healthcare. At the same time, I expect that there will be revisions soon. After all, we are still in the learning process and will therefore also learn from the mistakes we will make along the way."

This article was previously published in the February 2024 issue of ICT&health.