Proof of work via SHA256 hashing is really nice because you can validate it by machine. For instance a nifty project would be to get some remote attestation capable hardware like the
IBM 4758 cryptographic coprocessor often used by banks. Basically what's special about it is the hardware itself is exceptionally difficult to tamper with, and additionally IBM includes a mechanism called remote attestation where the hardware will tell you what software is running on it. Since these co-processors are used for many, many different purposes IBM can't release hardware that lies without damaging a significant amount of trust in them.
So, what you would do is write a very small, very simple piece of code that implements the Bitcoin block hashing algorithm. What this code would do is accept encrypted messages from anyone, either the query "What's the legit block chain?" or the statement "Here is the next block in the chain" Since the messages are encrypted the operator of the service can't prevent someone from telling the hardware about the best known chain, so anyone making a query asking what the chain is can be pretty sure that the response is accurate. The existence of this service would allow others to use it to bootstrap their own clients without needing to know any honest nodes at all.(1)
Smart Property is a good example where this service would be useful. Additionally it could argument or replace the checkpoint mechanism.
The problem with Ripple-style consensus is stuff like the above just can't be done because maintaining a list of public keys associated with trusted entities is fundamentally a task that only humans can do. For Ripple human consensus is probably a reasonable idea - Ripple depends on human evaluation of trust relationships anyway - but applying that concept to Bitcoin would turn it into something very different than it is now.
If someone has enough hashing power to rewrite every single blocks up to block #1, the IBM 4758 will still accept it as the legitimate chain but I don't think that's what we want. Human intervention is unavoidable in such circumstances. The question is, to what extent we could tolerate a blockchain rewrite? 6? 100? 2014? 210000? There must be a cut-off point where human intervention is needed.
The bitcoin wiki (
https://en.bitcoin.it/wiki/Contingency_plans#Many_historical_blocks_replaced) suggests that 6+ block rewrite is unacceptable and warrants taking down the network for more than a week to fix it manually. If we believe it is the right way to handle massive blockchain rewrite, why couldn't we consistently implement checkpoints based on human consensus?
As the IBM4758 is programmable, the list of validators could be updated manually. With remote attestation, it is not possible to inject malicious validators to the system.
If people are really uncomfortable with the concept of "elites", we could restrict it the "miners" and "riches". The identity of miners and riches could be determined automatically through the blockchain and no human intervention is needed. (With miners using the consensus scheme would not undermine the security, see below)
It also isn't a given that it would make Bitcoin any more secure either: if miners use this consensus scheme too, then by breaking the consensus you can either re-direct hashing power to your new, illegitimate chain, or failing that, turn the hashing power off to make a 51% attack easier. For non-miners consensus can help, but only in the sense that the consensus is warning you something is wrong, so you shouldn't trust transactions for now until we figure out what is wrong. Bitcoin already has a primitive version of that with the alert system anyway.
As the automatic checkpoints are for blocks with 6+ confirmations, someone may fork the chain up to 5 blocks. In this case, people will only accept payments with at least 6 confirmations. Trying to replace a block with 6+ confirmation is not possible because all honest mining and non-mining nodes will treat the original block like a hard-coded checkpoint. Although an evil miner may try to extend his own illegitimate chain, no one would accept it: once consensus is made, it is irreversible without human intervention.
For the alert system, I think it is controlled by the devs. As they are not constantly monitoring the network, an automatic warning system would help. Also depending to much on the devs during emergency is just another single point of failure.