This plays to your suggestion of emotional bonding being a key need for AI assistant's.
Though having it called 'Google assistant' seems to break that connection, though imagine something is in the pipeline being tested and analysed over and over to see what works with AI connecting with a user.
It's probably best left to a third party to skin the Google assistant with soft skills. A small startup can afford to gamble on these things. Google (or Viv, Facebook, Cortana et al) need only provide the underlying capability, in such a way that they can wash their hands of it should it invoke a public backlash.
Just so long as they're collecting the data they need to ultimately know you better than you know yourself.
If someone can build something you grow empathy and compassion for it will most certainly be a winner.
Google has had so many failures in the calendar/blog/assistant space because Google doesn't understand ordinary people.
It's an elite company full of elite engineers running elite engineering projects (and a lot of search). It has no internal cultural reference point for average-IQ non-elite users.
This is why Google has so bad at productising interesting ideas that Google didn't invent, and which have to deal with real competition.
The software works, mostly. But users don't feel there's anything special or unique about the experience.
You could argue - reasonably - this is all about perception management. Apple does well because it sells its products as stylish and magical. Even if the technological reality is mediocre, Apple has consistently managed to create a brand that is good at making technology seem humanised.
Google has never understood why that kind of marketing is necessary. The technology is good, it should speak for itself. If customers don't get it - no one understands why.
Now with AI, the pitch is out of key for the same reasons. Most people don't want something that anticipates their intentions, because that's just creepy. Even if it works well, it creates an experience of diminished control and empowerment.
It's like someone saying "Trust me" when they're promising you something new, but they've never given you a reason to. Does the AI know you well enough to make the right decisions? Why would it?
Unless Google can dramatise that it does, reliably, they're not going anywhere with this one, no matter how good the technology really is.
It's a technical challenge now, but in a few years making a good AI will be high art. By a few years I could mean 3 years or 23 years. There's the possibility that dressing an AI assistant up in a personality that asks 'how are you feeling?' could lead us into an uncanny valley that takes a long time to cross.