Soren Brunak is a Professor of Disease Systems Biology at the University of Copenhagen and Professor of Bioinformatics at the Technical University of Denmark. He is also Research Director at the Novo Nordisk Foundation Center for Protein Research at the University of Copenhagen Medical School and Medical Informatics Officer at Rigshospitalet, the largest public and teaching hospital in Copenhagen.
Dear Soren,
What could be achieved if there was a public or nonprofit AI effort with the same scale and level of funding as the current large private efforts? What would be the benefits for society?
I think one of the best opportunities would be to work with population-wide clinical data, because we would like to cover all diseases, and we would like to understand how they interact. Collecting population-wide data is of course a major challenge but in Denmark we have had a model, for close to 50 years, of keeping a lot of health data and also of having them in a linkable form. This system also allows us to have assurances in place about the correct origin of the data, to make sure that they are not made up and that they correspond to real persons.
We are also trying to link clinical data with known medical knowledge by mining full-length scientific papers. Extracting data from electronic records can help find the exposures and risk factors for diseases. We are now planning to use LLMs to process all this text data, including the side effects information from clinical trial reports and from drug package inserts. A challenge in this type of large-scale analysis is going from association to causality. We must be aware of the limitations of LLMs, including the so called "hallucinations", but we should not forget that doctors and nurses can also misinterpret symptoms or record them incompletely.
We are developing several AI tools with potential clinical applications, for example for Intensive Care Unit patients. Replication of data is an important step in validation, even if we must consider that diseases might vary in some respects among different countries. There might be differences in exposures, disease frequency and genetic background. Eventually a clinical trial would decide whether a tool is useful or not.
I feel that the explainability component in many of these AI methods could be really useful for clinicians. It is difficult for a clinician when you have an intensive care patient with 10 different problems and then you would like to rank them, to identify what is really taking the patients towards death, and sort those features according to whether you could intervene on them or not. It is also true that trained clinicians often develop effective intuitions about what works, even if they cannot explicitly explain them. We might be setting up some demands for the algorithms that clinicians might not be able to satisfy. As we all know, humans can also have conflicts of interests of many different types, not only financial but also related to their academic career and to high-profile publication ambitions. We all have a conflict of some type.
We are encouraging researchers at different career stages to share ideas about complex science problems that could benefit from a large-scale AI effort. We found that motivation and recognition could be provided if you and other well-known scientists were willing to talk to people that suggest the best ideas. You would be the judge and decide if any idea is for you deserving of attention. Any scientist selected might receive advice but could also be a potential collaborator. Many ideas will be produced, and society will take notice. Would you be willing to talk to any of these scientists?
Of course. We all get wiser by interacting. It might also help recruiting machine learning people to academia, given the current competition from industry for these types of experts. This is not to say that industry collaborations might not be useful, I have had many fruitful ones over the years. There is a necessary division of labor in the medical field between academia and industry.