I think many Apple users would be happy if only Siri could answer factual questions objectively and follow a conversation, being happy with a minimum of privacy collection like current location, name of spouse for messaging, etc.
While some may prefer a knowledge about you or you via people like you, as in, "Please recommend me a great movie", Siri is currently not in a stage where she may tell you stuff like "How do I mix a White Russian?" Google Assistant give me 6 steps with a photo and a follow up question "What about Screwdriver?" Siri gets utterly wrecked in the knowledge based questions.
I think the main problem is not privacy-related, but knowledge base-related. Google is building upon a fricking huge search engine via a knowledge graph and the sky is the limit for how well an AI can do. Apple is building on what, a shut down Ping social network, Apple Music listening habits, Wolfram Alpha hopes if all else fails, and a sparse Bing Search API if that failed too?
>While some may prefer a knowledge about you or you via people like you, as in, "Please recommend me a great movie". . .
Not to mention that Apple positioned itself as a brand for people who "Think Different." I don't think people who got taken in by that messaging would be attracted to the prospect of services that can more efficiently pigeonhole them.
I thought Apple figured out a way to create unique IDs and profiles that can't be traced back to the individual person? So they can still find people like you for analytical purposes to tune their models.
Speech recognition and natural language understanding.
If you watched the HomePod reveal you'd know that it sends your speech to the cloud for understanding. Which seems like a pretty clear admission that this can't all be done on-device easily.
The simplest example would be speech recognition. With the exception of Keyword Spotting Systems like Hey Siri or OK Google, it's currently practically impossible to implement larger vocabulary speech recognition on device.
Google will happily permit you to download trained models of <100MB to your phone via Google Translate, which permit not only offline vocabulary recognition of a remarkably wide range, but will also then translate that input into another language (also offline).
I believe the restrictions imposed by keyword-spotting have more to do with the always-listening and/or power-efficient nature of the task, rather than the restrictions on the legibility of offline speech recognition.
I'm pretty sure larger vocabulary local speech recognition on devices with less computing power than modern smartphones has existed (products for continuous speech recognition on PC go back to at least Dragon Naturally Speaking in 1997.)