I really hope to see the day there is an actual developed AI behind Google Home and the likes.
At the moment there are hundreds of hours of engineering work just to make it understand a single new sentence or command, which are basically predefined, coded queries that have their pre-defined answers/scripted clicks added to the database. The "human-like" texts are also pre-coded and scripted, written by Google engineers. Each new sentence that it understands takes a lot of voice samples for each way a person can ask to receive that given answer, and then QA. Everything else that was "not pre-scripted" is just a redirection to Google, where the voice recognition just inputs your words into a Google search. It's very primitive and ages from being anything better than that. There is nothing smart about Google home - from my perspective, it is unimpressive as is, very time consuming to improve, and the whole engine is just very basic code with a database of pre-defined actions as answers and a 'work-in-progress' voice recognition that is being marketed as a smart assistant. Considering new tricks are just new scripts, it hardly has anything to do with the AI we've been waiting for.
And then the "machine learning" thing - the only thing it "learns" is related to voice recognition, where it basically extrapolates the thousands of real-life voice samples it's fed into understanding the same words pronounced by other people who might sound a bit different. It also tries to filters the results by trying to primitively narrow it down to the ones you are most likely to ask for based on your location history, places most frequently chosen in the area, etc.
Hopefully the popularity leads to continuous development of machine learning and real AI engines that might pop up someday in the future. If Google Home is the very beginning of that path, I am happy that it's popular. But at the moment there is no AI there and the machine learning behind it is not even in its infancy - it's barely there at all. It's basically the "Hello World" of machine learning. After it's light years from the very beginnings that we see today, we might start talking about useful assistants and having something genuinely cool, at least in my book.