A bat and a ball costs $1.10.
A bat costs a dollar more than a ball. 
How much does a ball cost?

Not following how hardware mining vs stake mining answers that any differently.

Sure but I don't get what that has to do with distributed consensus systems.

In a PoS system, there's no tether to physical reality.  There's no one in the room screaming:

"A ball costs $0.05, a bat costs a dollar more ($1.05) so a bat and a ball costs $0.05 + $1.05 = $1.10 you idiots!  You all fell victim to your cognitive bias!!"

PoS is an echo chamber like Cypherdoc said, and although consensus might be achieved, it may have no bearing on reality.  PoW is the tether to the physical world.  PoW mining is the objective guy in the room "proving mathematically" with chalk and a blackboard that a ball costs $0.05 until everyone is forced to concede!

But why would they all agree when doing so would destroy the integrity of their stake? As we saw recently  getting them to agree on such a change is near impossible and will only get more difficult as the system gets bigger any many forgers are companies that have their businesses built on top of the system. History can be rewritten with pow too, it happened before.

Are you saying that AI will never be possible? Because that is a self referencing system.

POW is a better and more fair method of distributing new coins, but that is all.

Once bitcoin is widely distributed and the majority of block rewards comes from fees and not minting, then switching to a POS system seems fine.

Both systems are identical on how security is created, they only differ in who gets to vote. In POW votes are allocated based on how much electricity one consumes (assuming everyone has access to the same technology which will be true after ASICs settle down). In POS votes are allocated based on the number of coins held. The concepts of a voting based system are the same.

BTW bitcoin already did a rollback in March 2013, yes it was to correct a bug but the devs looked at the situation and asked miners to abandon the longest chain and build a shorter one. This is possible with POW too... Bitcoin is still based on the laws of men (not thermodynamics), its just that bitcoin aligns the interests of the "voters" in a way to create a rules based system.

So an entity sells their stake and then hopes that the rest of the network will agree to fork it back for them? Highly unlikely. If this fork, which would have only changed the outcome of a single known malicious transaction was unable to pass, then I can't imagine how a fork like that would ever pass.

PoS can make that promise.. in order to be able to rewrite history, a malicious user would have to buy well over 50% of the currency.. to do that and then fuck the system is as you say, uneconomical..

regardless, even if you are correct.. soon the roolback or reorg wont be possible.. transparent forging will soon be fully in place and when it is, you would need over 90%/95% of forging power to do a roll back or reorg and that would be virtually, or economically, impossible... or so unlikely that its enough... the same way a 51% attack is so unlikely it is enough.
[/quote]

Again, Vitalik has been doing a TON of research on this topic and look what he found: https://bitsharestalk.org/index.php?topic=6990

The Nxt community was debating "rewriting history" to rollback the theft from Bter!  The users become the attackers, and in their echo chamber they decide to make up whatever truth appears convenient at the time.  If you don't see this, then you don't want to see it.  

On the topic of a single malicious cabal attacking the network, the cabal needs only 51% of what is currently staked, which may be significantly less than 51% of the money supply.  Furthermore, the cabal can attack with coins they once had but have since sold.  This is the nothing-at-stake problem (aka the history-rewrite problem).  This can be prevented with checkpoints or if the developers sign blocks as valid, but then your decentralized currency is no longer decentralized!  We saw this with the viacoin roll back.

Lastly, the idea that some special sauce known as "transparent forging" will require 90% of the critical resource to attack is a highly dubious claim.  I strongly suspect that the following PoX Theorem holds:

A malicious user controlling 51% of the critical resource X in a PoX distributed consensus system can attack the network.