Earlier quoted context omitted.
Mass surveillance is bad, but I think there are versions of it that are far less bad than others. Apple's proposed solution would have theoretically only reported cases that were much more than likely to be already known instances of CSAM (i.e. not pictures of your kids), and if nothing else is reported, can we say that they were really surveilled? In some very strict sense, yes, but in terms of outcomes, no.
ok, and in theory, with new generative algorithms, do you think it's still ok? Suppose apple implements this, suppose someone finds a way to generate meme images that can trigger apple's algorithm(but human can't see anything wrong), suppose that someone wants to harm you and sends you a bunch of memes and you save them. What will happen? Or what does happen if somebody is using generative algorithm to create csam li…
I agree that the basic idea of scanning on device for CSAM has a lot of issues and should not be implemented. What I think was missing from the discourse was an actual look at what Apple were suggesting, in terms of technical specifics, and why that would be well designed to not suffer from these problems.