Most of the world was wowed by the Google Duplex demo, where their system was able to cold-call a hairdresser and make an appointment with her, with the hairdresser unaware she was talking to an AI. The system included human speech mannerisms and the ability to respond to the random phrases the hairdresser through back.

Since the dawn of AI with Alan Turing, being able to fool people into thinking an AI was a human has been one of the holy grail tests people want a system to pass. It was not explained very well, but the recent Turing movie was given the title "The Imitation Game" because that sort of test -- imitating a human to the point you can't tell -- was the one that Turing favoured. We don't call it the Imitation Game, we call it the Turing Test.

So naturally Google was proud and the computer science crowd was impressed. Some however, immediately picked up on the creepy aspect of this being used as a real product. Passing a Turing Test is a great demo, but using that tool to actively fool humans is another story. Some even felt it was unethical to even do it as an experiment, and also illegal in California due to conversation recording rules.

I hate it when computers pretend to be people myself. I hate chatbots. I even hate it when I get a form email that starts "Dear Brad." First of all, since real people don't write email that way any more, it is a clue that it's a robot. But secondly, the ability to substitute in a name in a mailing list tool was cool in the 1970s but does not impress anybody today. In fact, I would prefer they wrote "Dear Customer" because I want to really understand my relationship with the sender. They are not on a first name basis with me.

Others love chatbots. They know they are chatbots but still love them. Sometimes literally -- there are many documented cases of people reporting strong emotional bonds to chatbots. I'm in a minority, I am told. Give me a command line or an easy to guess natural language syntax any day.

Google has talked about enabling this function with their assistant, so it will make calls for you. That bothers people. They don't like the idea of being called by a robot. We all hate it when called by a recording, and the laws that now forbid this are called laws against "robocalls" -- which should give a hint.

Google has since declared the system will always identify itself and what it is.

But there is an easy path out for Google. Market this, not as a service for the caller, but as a service for the business.

Even 25 years into the web, lots of businesses don't have a web site where you can do many things with the business, including booking appointments. If they have it, they don't do it in a standard API to make it easy for electronic assistants and aggregators to use it. But they all can do everything on the phone. Google can offer the business a way to be "on the web" without having to change any software, or even train their employees. They can tell the business to enable Duplex and then anybody using Google assistant or other tools can suddenly book things over the web or with a mobile app.

Most businesses would say yes. Everybody wants more business, and wants it to be easy for customers to make bookings. As long as the tool does it well, makes few mistakes, and doesn't call so much that it can't be handled. (For example, Duplex supports calling and asking what holiday hours are. You don't want such a call happening every time somebody looks up the place online and clicks to see the hours. It is OK if the first time in the season, or for a specific day, it calls and then remembers, though.

So if Google wants to bring the old world onto the web, this could be a good tool for that. If it becomes a tool to annoy and impersonate, that will be the wrong direction. One big fear will be its use for real robocalling. The robocallers will not just play you a recording. They will even make software that uses the voice of their human sales agent so the robocaller begins the call, qualifies the target and the human sales agent is brought in seamlessly.

The laws already forbid robocalling, and if for some reason they don't forbid this, they should be amended to require disclosure of what is calling you. This may not be very useful though. The rise of cheap VoIP now means most robocalls just come from outside your country, and your laws do not apply.

However, regardless of how Google acts, this technology will soon be developed by others and will be used to deceive. Science Fiction stories have frequently featured the idea of an AI personal assistant who pretends to be you -- it duplicates your voice, face (on video chat) and some of your knowledge to deal with incoming callers, and in rarer cases outgoing calls. The funny thing is this SF didn't predict the way we would switch to using other media (like texting) to set up conversations. Voice mail and IVRs are on the way out, not in. I hope in time that we realize that we have all sorts of ways to communicate, and our tools should help us always use the right one. We don't have that today. People send texts that are not at all urgent and should have been an e-mail. They have long text chat conversations which really should be a phone call. In time, I hope that we get to a place where we're never trying to fake one medium with another. Duplex uses a phone call to do what should have been a web API. That should be a temporary problem.