What needs to happen is a company needs to come along and create AI that is trained off of generalized information... some kind of socially accepted public data set... then the trained core is sold as a seed to individuals who then feed it their personal data.
It'll be the equivalent of buying an AI "teenager" and slowly training them to be an "adult".
As sensors become richer and the data becomes more valuable to the ML, consumers are becoming more aware of their privacy.
That means to get to a 'good seam' in the future instead of trawling through trash, you're going to have to convince millions of people their interests won't be affected.
That means in time there is an opportunity for a Google-killer with a different business model not based on using the raw data OR using by the use of intelligent agents. Google goes down because its stakeholders are contingent on getting to the raw data.
ML relies on large data-sets and if anyone tried to release a personal device it simply wouldn't even work, let alone compete with the mass surveillance google/ms/amazon are bringing to bear.
Unless the state-of-the-art in AI suddenly morphs, we seem to be stuck between giving up our privacy or having vaguely intelligent AI.
I personally fall heavily on the privacy side of stuff, but I can see the intellectual and commercial appeal of pretending it doesn't matter in order to get there.