* Let people know that you will be late to a meeting (in case they don't check their text messages or email).
* Conduct over-the-phone surveys.
* Automated surveys for science (this one was suggested by cocodilux on another forum).
* Cancel a service that you no longer use (but get bills for anyways, because you have it set on auto-pay).
* Haggle with a group of people on when's the best time to meet up for group work (the bot would call everyone, ask about availability, until the time is worked out). This is basically what the company X.AI is working on, except they do it with text.
* Call and remind people about something. For example, reminding elderly people who prefer to use the phone that they should take their medicine.
* Call and ask people for information you are too embarrassed to ask (maybe Google has a list of 10 types of embarrassing conversations it can handle for you).
* Apologize on your behalf (maybe there are 10 standard apology conversations). Sincere apologies should probably be personally delivered; but there may be some cases where automating the process is acceptable.
* Contact a bunch of places to see if they found a lost item (e.g. you could have Google contact the last 5 places you visited; and it would have records of where you were, based on the GPS signal).
* Call and ask if a store has a particular item in stock, that you want to buy it right then.
Perhaps training the system to use webpages will have an even bigger impact, as so many services and businesses are online these days. If you can just ask to get things done in natural language, Assistant will go and use an API when it can -- and if it can't, it will just use the user interface on a web-browser. So, you could say, "Cancel my subscription to X" and it would maybe go to the website, and use the mouse and keyboard, enter your information, and then cancel it for you. As the complexity of things it could do grows, you could issue commands like, "Go find a list of all the people who attended that seminar, and send them an email that I am greateful that they came."
Incidentally, a couple years ago, right after I wrote a post on the future of computers using web browsers to automate tasks (I believe Yuli Ban posted it here on this forum), coincidentally OpenAI started a project to do that very thing. Andrej Karpathy was behind it. He now is the head of AI at TESLA. The other day he wrote a Tweet:
Duplexes talking to duplexes would be an amusing use of existing (human) interfaces but by AIs. Just like autonomous cars use existing roads/signs for humans, or how our "world of bits" AI at OpenAI used simulated keyboard/mouse events to interact with web pages. Very amusing.