Credit where credit is due! The following scheme was described by Ben Laurie years ago:

The "central authority" consists of a set of distributed (and preferably independent) servers which keep track of the ledger of balances (a Merkle hash tree) using a Byzantine-fault-tolerant consensus protocol. They enforce constraints on the system (e.g., coin creation and distribution---no need to do it manually or premine) and blacklist dishonest or malfunctioning servers.

All the honest servers will agree to the same transaction log (of course a severe netsplit will stop transactions from committing). The clients need to come preconfigured with a list of servers (which can be updated by the network). So, unless something goes very, very wrong, there is no need for the user to worry about deciding whom servers to trust.

There are clearly some details to fill in to specify a complete implementation, scalability issues, etc., but Laurie's distributed currency uses established ingredients that are mathematically proved to work and gives you transaction confirmation in seconds (no need to "mine"). It could also be used for Bitcoin and other currency transfer/exchange (there seems to be a need for this anyway) as well as more advanced uses.

Now, I indeed remember the original ripple web-of-trust proposal being as gmaxwell described. The "new" system seems to be something else, and I echo people's concerns expressed in this thread and do not see how they have been resolved to any satisfaction. It may be unintentional on the part of the ripple developers, but something seems fishy or opaque about it (at least at this point in time), and it looks like a commercial rather than a community project.