He is solving one problem by trying to directly accessing the data, and then causing another problem because he has to deal with autovivification.
The code becomes larger and harder to understand. I am also willing to bet that this sort of 'check' would be used in more than one place in his program. It should be abstracted out as method and named so that it makes senses in the context that it is being used.
Another reason to avoid directly accessing deeply nested hashes is when you need to change the data structure. If you are accessing values directly, you end up changing a lot of code to use the new data structure. If it is hidden behind a method, then you only need to change the method to fix the program as a whole.