Here is an example, I would for instance use Google Maps for Navigation but Google or any other third party would have no idea where I am going.
I used it in the first company I worked for and it works beautifully.
A) and B) work but they are not as effective as homomophic encryption.
You don’t need homomorphic encryption for this, just local route processing. In the case of car data, the auto companies aren’t doing any useful processing of the data for the user. Homomorphic encryption is irrelevant.
Homomorphic Encryption reduces the breadth of computations that can be ran on the gathered data, by making it inaccessible outside of the specific homomorphic scheme that was chosen. So yes, in that sense it cannot be used arbitrarily.
However, the results, i.e. knowledge derived, from the chosen computations can still be shared arbitrarily, which IMO is a much greater issue, as the need of the result sharing will inform the computations that can be done within the scheme.
Who defines the computations? Surely not the users, and lacking regulations, also surely not regulatory bodies.
For example the USG is forbidden from collecting communications from US citizens, but that does not keep it from buying this information from private domestic sources or from other governments.
Why is everyone so quick to say 'well, they are getting away with it, might as well let them' instead of trying to use our processes for the purposes which they were designed?
Homomorphic encryption is where you can compute on the encrypted data without ever decrypting it.
Logically, it sounds like a pipe dream to me, but apparently it's a thing.
The truth of the matter may be something other. Life is not always logical.
We can only do a limited set of operations homomorphically. Moreover, it’s more power intensive than conventional computation. In most cases, local computation is the more effective (and secure) solution.