It's true that RAID is not a backup, but backup strategies are rarely any good if they back up data "instantly", which is what redundancy provides. So you're stuck with a catch 22. If you back up instantly (without versioning), you may backup bad data. Boooo! If you backup, say, daily, you still stand to lose a days work when your array fails. Boooo!
RAID should increase reliability, not decrease it. Any RAID solution that results in a net increase in data loss is not a good solution.
That's a catch 22 that can be minimized or even avoided altogether. My main drobo backup is automatically versioned and backed up every 12 hours (to a single external drive, since I don't need to back up everything on my drobo). I've got older backups dating from like a week (work-related) to half a year or more old (for the important docs that don't change that I should keep around regardless...like old accounting/tax stuff) stashed around my home, online, and in a safe deposit box at my bank. And of course occasionally I test whether or not my backup is actually backing up the things I want and that I can successfully restore from it.
If I back up bad data, I have older but good data to recover from. If my drobo catastrophically fails, at the worst I've lost data that is probably not going to be difficult to replicate/replace because of how new it is. I don't have to start ripping my hair out at the idea of having lost an incredible amount of data ranging from irreplaceable photos of my life to work and more.
In the ideal case, raid should increase reliability, and they do, only for one definition of reliability. But freak accidents and poor design happen, and it's silly to rely only on a raid for reliability.
RAID should increase reliability, not decrease it. Any RAID solution that results in a net increase in data loss is not a good solution.