I don't know that original title was but a quick look doesn't show the word "language" used in the article. "The language of the immune system" really seems like editorializing.
In fact, the title as I currently see it: "Modelling the lanuage of the immune system with machine learning (first steps)" seems like a pretty strong title abuse from start to finish. The title of repository is: "Statistical classifiers for diagnosing disease from immune repertoires".
This is obviously a rubbish comment, but nevertheless it raises an interesting point. Deep learning has entered a phase where, due to the excellent tooling (PyTorch, TensorFlow etc) and freely available data, there is a massive ingress of new poeple into the field. This has had a variety of effects, many of which are not necessarily beneficial to progress. There is so much noise and hype that it is difficult to do good work in the first place, and then be recognised for it. Unsurprisingly, rumours about massive salaries at FAANG, everyone trying to start AI companies, and a torrent of dubious quality research flooding conferences retard real progress.
I suspect that there is some optimal mix of making a field accessible and attracting good people to it. If it is too inaccessible, then progress is too slow. It it is too accessible, then it gets flooded. Most scientific fields exist on the inaccessible spectrum. It is relevant to note that many amazing discoveries (including the foundations of deep learning for example) were made when the field was obscure.
So to answer your comment, a scientific field doesn't owe you anything, and in fact, is probably better served by not making it too easy for you to get involved.
I'm actually happy if basic and understandable classifiers are used, this might make the model tell us something about the underlying biological principles.
