I think this is a general problem with a lot of academic research. When someone develops a new technique for solving some abstract problem (a new graph algorit

I think this is a general problem with a lot of academic research. When someone develops a new technique for solving some abstract problem (a new graph algorithm, a new type of ML model etc) they want to demonstrate that it has practical applications. What often happens is that they take some problem domain they aren't experts on (their expertise is in maths, comp sci etc), apply their methods to the best of their knowledge and publish the results. Since such academics read each other's papers, this leads to a gradual divergence between what's in the literature from what's actually useful. I think one way to tackle this is by forming interdisciplinary teams. For example, I work at an industrial research lab on AI in healthcare, and our project team primarily consists of various clinical specialties. ML research and engineering are around 20% of the overall team.