I think Google's concern would be that it would generate annoyance that might not be warranted. Like, if I call a customer service line and get an AVR system, e

I think Google's concern would be that it would generate annoyance that might not be warranted. Like, if I call a customer service line and get an AVR system, even if the system is perfectly capable of helping me, I just hate AVR systems (justifiably or not) and start yelling at it to give me a representative until it does. If the AVR system were so compelling that it just pretended to be a human and I couldn't tell the difference, presumably I'd be just as satisfied as if I had gotten an actual human, and could skip the annoyance. Similarly, I think if this thing can handle the call as well as a human, it's to its advantage in terms of not frustrating the person on the other end of the phone to not self-identify, so that they perceive it as just like any other appointment call (because it is) rather than triggering an "ugh damnn it, another Google robocall" which will then prime people to feel like they have to speak to it a certain way to get it to behave, even if that demonstrably isn't actually necessary, and generate ill will towards the company. On balance, maybe they should self-identify anyway -- "if this thing can handle the call as well as a human" is a pretty big "if" after all -- but I can see why they'd want not to.