Three success factors for AI (in healthcare)
Data quality, autonomy, and utilizing the benefits of people and machines
Hey there,
Artificial intelligence is playing an increasingly important role in healthcare.
Smart software and apps analyze data to estimate whether a patient can be discharged from intensive care, help radiologists assess scans and can calculate the risk of Alzheimer's disease based on a person's voice.
In this newsletter I focus on three success factors when implementing AI in healthcare.
But it’s also very relevant if you are working in another field (with AI).
Enjoy reading!
Peter
P.S. I made this video about examples of AI in healthcare (16 minutes, in Dutch):
Success factors
The potential of artificial intelligence in healthcare is enormous. Smart software helps to quickly convert recorded text into reports, estimates the chance that a patient will develop an infection and helps a novice to detect Parkinson's disease more quickly. But where should you start?
Of course, there are plenty of comments to be made about the use of AI in healthcare. But apart from that: if you want to get started with it, what are some smart tips and success factors? Here is a selection of 3 factors:
Power of people and machines
Autonomy and job satisfaction
Data quality
1. Power of people and machines
Certain tasks that machines excel at are difficult for humans. And vice versa. This phenomenon is referred to as Moravec's Paradox.
The best example is driving a car. It has been said for decades that fully autonomous vehicles are almost here. However, driving a car requires two things:
a comprehensive mental model of the world;
the ability to interpret unexpected situations.
Algorithms are not good at this (although they are getting better). If a task contains these elements, then it can be done even better by a human.
AI can do other tasks better than a human. Consider assessing scans by a radiologist, pathologist or other medical specialist. A specialist cannot possibly compare a hundred images in a few seconds. Let alone that he or she can compare that directly with the scans of the same patient from a year ago.
That is not a problem for an algorithm.
2. Autonomy and job satisfaction
AI applications have added value if the quality of care improves, if healthcare costs are reduced, if patients heal faster and if the work of healthcare professionals becomes more interesting, better and more fun.
Radiologist Paul Algra is a well-known advocate of AI in healthcare in the Netherlands. He states in an interview that:
'the use of artificial intelligence has made my work more interesting and improved the quality of the work.'
Research shows that autonomy is an important factor in the job satisfaction of professionals. If AI is too controlling and dominant in the work of a doctor or nurse, autonomy and job satisfaction decrease.
If job satisfaction decreases, you run the risk that healthcare professionals will not use the AI software or will (secretly) sabotage it. So it is smart to look for how AI can make work more interesting (in the words of doctor Algra).
3. Data quality
Machine Learning works only as well as the labeled data it is given to train. Therefore, it is important to accurately collect and label the training data.
Take, for example, the skin apps for recognizing melanoma. These are trained with datasets of predominantly white people.
A phenomenon that can occur is convenience sampling. This means that the AI developers have trained the algorithm with data that strengthens the functioning of the model. Unwilling data on which the model scores less well can, for example, be deliberately left out of the training set.
Another element is that AI programs that do great at one hospital usually turn out to perform much worse at another hospital. Eric Topol pointed this out in a major academic review article. The differences in patient population and probably the metadata, due to the method of registration, play a role in this.
Are you curious about the other 2 success factors? Then read my Dutch article about AI in healthcare.
Deep dive
Articles, books, podcasts, videos, documentaries and more on this theme.
1. READ / The book Smart Until It's Dumb by Emmanuel Maggiori is about the development of artificial intelligence algorithms and the (intentional) mistakes that can be made.
3. WATCH / There are plenty of television series about healthcare, from Grey's Anatomy to House. Me and my girlfriend Susan enjoyed watching the sitcom Srubs a few years ago (8.4 on IMDb).
This scene is about a robot that comes to tell patients the bad news (90 seconds):
🙏 Thank you for reading
This newsletter is free, but not cheap to make.
You can help me in a number of ways: forward this newsletter to someone who likes it, subscribe to my YouTube-channel, hire me to speak, for a workshop, or for a webinar.
Interested to hire me? Feel free to fill this form!
