Google has been polishing its phone assistant software to a fare-thee-well, to the point where its capacities are becoming both exciting and disturbing. Its new platform, "Google Duplex," can handle basic tasks, adapt to some variations on phone encounter scenarios, and (here comes the disturbing part) sounds very much like a human being! Duplex's capabilities are developed to the point that, in some but by no means all situations, it is claimed to be able to pass the Turing Test, that is to say: to demonstrate a machine's ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human.
Think I'm kidding? Watch the following, then you tell me
That is amazing.
At that restaurant call, it sounded like the AI was more intelligent than the human answering the call.
So far, I'm more excited than disturbed. It sounds like it won't be too long until talking to AI will be a more productive conversation that I've had talking to many humans.
What specifically disturbs you, Loren?
What disturbs me? Wondering whether I'm talking to a machine or a human being. Whether what I say and how I say it is being used in further AI research without my knowledge or consent.
Most of all, wondering whether or not this is the leading edge of the obsolescence of Homo sapiens.
Loren, what can a human handle that a programmed AI system can not handle?
A change of context.
For instance, if making an appointment, try a small or a large change.
A small change: "I want to talk to the manager."
A large change: "Is there a nearby sporting goods store?"
The possibilities are endless.
The question you have to ask is: this "programmed AI system" – what is the extent of its programming? What is the environment it is prepared to deal with and will it admit that it doesn't know when the parameters of its programming are exceeded? "I'm sorry, sir / ma'am, but I have no information about that topic" is the kind of response I would expect, and were I a programmer, I would probably include. The scenarios represented by the video above have likely been repeated thousands if not millions of times, and with rare exceptions, are likely pretty routine and therefore reasonably easy to program. Also keep in mind, the AI in these cases is represented by a customer and NOT the restaurant or other service. An AI operating in a restaurant or other service, fielding calls from customers who may be happy, sad, angry, or irrational I suspect would be far more difficult to accomplish.
Side note: the AI being able to admit to a lack of knowledge makes sense in this venue. Would that believers could do the same thing!