However, the reason these are not comparable is proof of work/investment. Presumably Duplex can or will soon be able to make millions of calls for any purpose whatsoever. Unlike a human being with a foreign accent perhaps or other alerting auditory factors Duplex massively reduces the cost of experimental manipulation.
One of the most successful hacking techniques is phishing. Its success is directly tied to how compelling it is at counterfeiting a legit communication.
And this does not even touch on larger institutional deception.
Being worried about Duplex in its current form is like being worried about Siri becoming conscious. Your concerns about malicious AI voices are valid, but that eventual technology will be nothing like Duplex in its current form, and it's still a long way off
The outcomes I speak of, which are nothing compared to all potential exploits, are possible with the existing Duplex technology, in that they do not happen it is only in Google's good gatekeeping. That is too thin a standard for me.
I suppose Duplex could maliciously attempt to book a bunch of fake reservations, but the technology behind it really can't do much more than that
Also, I'm not really sure what you mean by "Google's good gatekeeping". 3rd parties have no control over Duplex, other than initiating specific tasks through Google Assistant. If Duplex does anything malicious, it would be because Google went rogue, and there would obviously be massive legal and PR consequences for that.
Why? You presumably exchange text with chatbots all the time, and get mail/slack etc updates from your CI bot as well. Do they bother you?
you get phone calls / talk to people in shops who claim to have your interests in mind when they steer you to one product over another. Does that disturb you?