“For many people in 2024, AI was a kind of magical black box they did not understand. ChatGPT sometimes seemed like a real person, while it was a system that could only parrot people very well. The big companies behind many systems for AI were not at all open or transparent and had little regard for sustain- ability. People did not know how those systems worked. We tried to break that open over the past five years. In the case of our project about the use of AI in healthcare, we looked at how doctors can rely on AI? They can only do that if they can figure out exactly why AI comes to a certain conclusion, when it can explain why it says something.”
“If you go to the doctor now, in 2029, with certain symptoms that could potentially indicate cancer, AI will help with the diagnosis. AI can very quickly access different sources of information and recognise patterns. AI can even detect and classify tumours on a photograph better than humans, for example. The problem was that sometimes a human user did not recognise the classification. The AI should then be able to explain why it does classify something that way. It now says: ‘I know with 95 per cent certainty that this is a tumour and the 5 per cent I am not sure about, is because of this or that.’ That could be too few data points or an image that is not sharp enough. That black box had to open, the AI system had to be able to explain in a way that was recognisable to humans why it was saying something.”
Cats and dogs
“I always explain this using an AI that you give thousands of pictures of dogs and cats, after which the AI has to learn the differences between these animals so that it can recognise a dog or cat. When we see a dog or cat, we recognise it by its whiskers, its ears or whatever. But an AI might recognise a cat by the curve of its spine: something we humans do not do. It works the same way with tumours: maybe an AI will recognise a tumour by something we as humans have not thought about, let’s say: always the same ragged edge. If the AI can explain that to a doctor, the doctor will learn from that too.”
Treatment plan
“After making a diagnosis, the AI can further help with the treatment plan. Five years ago, such a plan was mainly put together based on a patient’s lifespan. Now we look at it differently: not only extension is important, but also quality of life. So adding more days to life, but also more life to days. That consideration is purely personal. It depends on the patient’s age, whether he has children or a partner, whether he likes bridge or cycling – and what he would like to keep doing. You can now feed all those elements into an AI system, and then the AI can calculate which treatment gives the most chance of the ‘best’ life.”
Personalised care
“Now, in 2029, as a patient, you engage with your doctor about a treatment plan. The doctor knows what you like in your life; the care is entirely personalised. Five years ago, you only sat at the table with the doctor and he would present the plan. Now you sit at the table with the doctor and a third entity: the supporting AI system, which can calculate scenarios. If we do this, the chance of you being able to keep playing bridge is 30 per cent. If we do this, that chance is 70 per cent. So there is an interaction with three pillars: the patient, the caregiver, and the AI. The patient and the caregiver together make decisions about that personalised care.” “AI will never replace a doctor in healthcare. It is always the doctor and the patient who are in charge, the AI supports. It is about the patient’s well-being and overall state of wellness. That sometimes required a switch from doctors, as they were trained to extend lifespans. But now it is much more about quality of life. AI plays a crucial role in that.”
PersOn consortium
More people are diagnosed with cancer every year. Often they will receive a standard treatment that does not pay attention to the patient’s expectations and wishes about the quality of the remainder of their life. The PersOn program analyses, using artificial intelligence, both the clinical information on treatment and follow-up as well as the possible impact on the patient’s quality of life, such that patient and health care practitioner can make a shared decision about personalised care.
PersOn website
