There have already been multiple incidents involving bugs in "smart contract" code being exploited to steal funds, one within the past few days[1]. A particularly nasty case involved something called "the DAO", which was an attempt to do something like a distributed VC fund, but had a bug which allowed any participant to steal all funds contributed. That one got backed out by a hard fork of the Ethereum blockchain, but that's obviously not a stable or scalable solution to dealing with these types of situations going forward.

[1] https://www.theregister.co.uk/2017/07/20/us30_million_below_...

Basically yes, and this is reasonable from a security perspective, otherwise every smart contract would basically have a trusted administrator.

But in many cases you do want some ability to upgrade the code, and then you simply program in that ability by creating a repointable proxy with whatever authorization logic you want.

Yes the block chain is by definition immutable. So no means to update the contract the only way if it is a ENORMOUS mistake would be to fork it: read about DAO hack FYI

>what happens if there is an error in the logic?

If you are a big name in the Ethereum community, they will soft fork Ethereum so that your mistake is rolled back.