Following the nicely put questions by Richard Cave, I propose here are some suggestions:
1) LOOK and CONTENT OF DOI:
Like other references (isbn, pubmed id, etc) the "look" should be data-base centred and thus include only serial number-letter. This secure approach does not preclude to include the year of submission in 4-digit code (2006).
Including name or e-mail would be by far less reliable. Just think about translation of Asiatic names...
This obviously not human-friendly look should not preclude the usage of author DOI. Indeed, it could be included in any address and easily accessible. Likewise, copying and pasting a phone number is not difficult !
Regarding the content, this should be adapted to needs. I guess that current data on name, affilation, address, phone, fax and e-mail would be enough to contact the authors.
2) SECURE AND ACCURACY:
The security and accuracy could be easily reached if a submission or creation of doi applies upon a submission of scientific data (manuscript, data submission into data base,..). This implies that submitting DOI for authors would be prospective only. This would like any scientific submission.
3) UPDATE THE DATA:
The authors themself would be responsible to update their data either through login or submission to data base authorities.
4) OPEN ID:
I guess this could be feasible.