> Just sounds like they've been diagnosing by symptom because they didn't know the actual etiology. Once it's known shouldn't they just change the diagnosis to be the new, more specific condition? Seems like diagnosis is still the goal?

Well, we don't know the aetiology (in most cases), so while restructuring diagnosis on the basis of aetiology is defensible as a long-term goal, we are nowhere being able to achieve that yet. The argument is, designing research around the current symptom-focused diagnostic categories is blocking progress in research. Two people with the same diagnostic label may have completely different aetiologies, two people with the same aetiology may have completely different labels, two people with the same aetiology one of them may have a label and the other may have none – the same underlying brain dysfunction may manifest with very different symptoms in different individuals, depending on its interaction with other biological factors, and their social environment. If we want to understand "what's really going on", it can be better to put everyone in the same sample (both diagnosed with various diagnoses and undiagnosed), then look for patterns in that big sample. I think this paper is a great example of doing that, with the help of machine learning – https://www.nature.com/articles/s41398-019-0631-2 – but it needs replication.