AI in healthcare: Potential, obstacles and the risk of shadow AI

How is AI used in healthcare today?

- Very different. Some researchers and clinicians are curious, but unsure, others are already very advanced users. Since I came to Karolinska Institutet in January 2020, my mission has been to map out who does what and how to support each other. It is clear that interest is growing rapidly, but also that the needs for support look very different.

What is the biggest obstacle?

- Getting from prototype to real use. Many come up with solutions that work well in tests and that everyone thinks seem promising. But when they are to be introduced in care, it will be a long drawn out process. It requires permission, certification and a clinical introduction, which takes time. The risk is that the solution is out of date already when it is ready.

Is that what is called the pilot graveyard?

- Yes. That's where projects end up that work, that have been tested and validated, but that still never start being used clinically. It's quite tragic. Care misses out on good ideas because the last step is not completed.

Why does that happen?

- Often because the introduction is not sufficiently anchored in the business. It is not enough that the technology works. The clinic must have time, resources and a clear need. Otherwise, the project gets bogged down just before it could do any good.

You also warn about shadow AI. What is it and how big is the problem?

- Probably bigger than many people think. Employees use AI tools at work without being approved by the employer. This could be medical records, patient information or medical images. It worries both management and lawyers, as it can violate confidentiality and regulations.

Why does Karolinska Institutet need a Center for AI Innovation and the initiative AI@KI?

- To gather skills that are otherwise scattered. We wanted to create a common entrance for KI's researchers, where they can get help with both methodology, law and what is actually permitted. In short: Come to us and we will help you.

How big is the potential for AI in healthcare?

- It is very large. AI is already used in image diagnostics, for example in X-ray, magnetic resonance and computed tomography. The technology can contribute to sharper images and better decision support. But it's also about control: how much you want to buy ready-made and how much you want to develop yourself to ensure that the solutions really meet the needs of healthcare.

What risks do you see?

- That the staff become too dependent on the systems. If AI services suddenly stop working, the clinician must still be able to do the job. Studies show that things we don't use regularly weaken. In healthcare, it can become a patient safety problem.