It can be very helpful when trying to figure out why one machine won't understand another, for instance.
You can put meta-data in for debugging without compromising anything, schema wise.
Or in the case of config files, there can be detailed instructions on what fields are what they should contain.
The thing about XML is that it strikes a sweet-spot between machine readable files and human readable files. (I can't believe I'm coming out as an XML apologist!)
If it were only "by machines for machines", we wouldn't consider JSON, YaML or XML as much, we'd all go for Protocol Buffers or Parquet or something.
