The article's description of a DOI is a bit vague. It isn't just a link. When you register a DOI, you have to register the metadata, which are then stored with your DOI registration agency (like crossref or datacite). e.g., 10.15783/C7F59T is a DOI for one of our datasets. If you follow http://doi.org/10.15783/C7F59T in your browser then you get taken to our data archive, but if that goes away then you can still query the metadata at datacite (http://data.datacite.org/10.15783/C7F59T)
In the second instance, we're as open as possible about it and discuss the shortcomings of the system! It's not perfect, but nothing is. We think it's the best solution.
Thank you for the clarification. It clears a up lot; it seems I gave the designers of the DOI too little credit. Your last question brings some food for thought. The naive answer is to "just decentralize it", but smarter people than I have been trying to tackle a similar issue with certificate authorities. Sometimes software isn't really the easiest solution in contrast to open dialogue and cooperation.
Most readers on the blog know all about DOIs so that section wasn't meant to describe everything about them. I welcome feedback though.
What happens if doi.org goes down? In the first instance you can use an alternative resolver, e.g. http://dx.doi.org/10.5555/12345678 can be resolved against http://dx.crossref.org/10.5555/12345678
In the second instance, we're as open as possible about it and discuss the shortcomings of the system! It's not perfect, but nothing is. We think it's the best solution.