(edit: see my reply to your other comment. I think that one's a bit better-structured to address the points you mention)
I don't think it's out of the question. Here are some good reasons why Facebook might want to structure their pipeline like this.
- Cellular Bandwidth is still expensive. They don't want the reputation of raising your data bill by uploading all your photos. Maybe the GPU cycles really are cheaper. After all, Facebook already uses a lot of battery anyway; what does a few million more cycles per image matter?
- Messenger on iOS gets a new release every week. It's not very difficult to make code changes. Not quite as easy as a webapp, but updating recognition models is a pain anyways, so it won't happen too often.
- Uploading photos without the user's consent is going to make Facebook look very bad, especially in certain countries like Germany or other parts of the EU. They already don't have a lot of social "privacy trust capital" as it is; they might not want to waste even more "trust capital" on such a silly application.
Compare this to Apple, for example, who loudly claims that even they can't read your encrypted iMessages. They do this to win privacy karma points. Sure, this claim may not be true, but Facebook can't even consider this kind of marketing if photos are indeed being uploaded. On the other hand, if all the recognition happens on the device, Facebook could win some much-needed privacy karma points back by loudly saying so! Why wouldn't Facebook want to be able to say this? It's within their grasp, too!
I agree; I don't think they're doing this. But my argument is much weaker: I only claim that it's possible, or perhaps even reasonable and desirable, to build their service like this.
I don't think it's out of the question. Here are some good reasons why Facebook might want to structure their pipeline like this.
- Cellular Bandwidth is still expensive. They don't want the reputation of raising your data bill by uploading all your photos. Maybe the GPU cycles really are cheaper. After all, Facebook already uses a lot of battery anyway; what does a few million more cycles per image matter?
- Messenger on iOS gets a new release every week. It's not very difficult to make code changes. Not quite as easy as a webapp, but updating recognition models is a pain anyways, so it won't happen too often.
- Uploading photos without the user's consent is going to make Facebook look very bad, especially in certain countries like Germany or other parts of the EU. They already don't have a lot of social "privacy trust capital" as it is; they might not want to waste even more "trust capital" on such a silly application.
Compare this to Apple, for example, who loudly claims that even they can't read your encrypted iMessages. They do this to win privacy karma points. Sure, this claim may not be true, but Facebook can't even consider this kind of marketing if photos are indeed being uploaded. On the other hand, if all the recognition happens on the device, Facebook could win some much-needed privacy karma points back by loudly saying so! Why wouldn't Facebook want to be able to say this? It's within their grasp, too!
I agree; I don't think they're doing this. But my argument is much weaker: I only claim that it's possible, or perhaps even reasonable and desirable, to build their service like this.