Captain DaFt once typed: "So far, the corporate definition of AI seems to be tending toward voice activation/speech recognition, data sorting algorithms, or just whatever software Facebook/Microsoft/Google/Oracle/etc. has being developed for them since 2010."
Good point, it's a bit like voice controlled automation, you could replace the voice input with a button and have it do the same function, less any voice specific contextual processing/output. The AI we are waiting for is a system with open-ended input and contextual output so fantastic, you'd think naked angel babies had brought it to you from another dimension. Right now the various so-called AIs of the world can do very little, but the voice recognition in most cases is good. Google Voice is my favorite example; it's doing voice to text translations (sometimes badly, but it tries) without any training on the user's part. Siri is nice, but you must train it to recognize your own voice. Good for security, a pain when you get a new device and have to talk to it before playing. In the end though, they do very little after the voice recognition is over with. I have spoken to Siri twice in my life and both times she failed and just did a keyword search. When I say; tell me how to do a screenshot, it should figure that out and just say "hit the home and power buttons, guy" but it could not figure that out. I get it; Siri does appointments, voice calling, and a couple of other things... things that are pre-built and crafted before hand. I need her to do more random things before I qualify her as a AI. Again, that's just voice input automation. Very little is being figured out by the "AI" backend. And when I can do those tasks faster without having to talk to a less than full featured assistant, that is telling.
Anyway, Google is on the right with this DIY effort. Just let us poke at the AI and do our own local customizations, feeds, displays, etc without having adverts sprayed at us, or not being able to cut off the upstream data slurping, if we feel like using it privately. I don't see any downsides to this little project, other than not having control of the hardware once their OS or service is running on it, and the aforementioned lack of real AI features in the return data from the call to the mama cloud AI API.