I watched the Duplex demo three times Wednesday, and what got me was not the booking. It was the mm-hmm. Google's assistant called a hair salon, asked for an appointment, and when the woman on the other end paused to check the book, the machine filled the silence the way a person does. Little verbal shrug. Perfectly placed.
The room at I/O laughed and cheered. It is a genuinely astonishing piece of engineering and I will not pretend otherwise.
The woman at the salon did not appear to know.
That is the part that has been sitting in my chest for two days, and by Thursday it was sitting in a lot of other people's too, because the backlash arrived fast and it was not about the technology.
Brand voice has always been a performance of personhood, and we have just been handed the first tool that performs it well enough to be mistaken for the real thing.
HAL is the reference here, and I mean it structurally, not as decoration. What makes the computer in 2001 frightening is not that it is cold. It is that it is warm. It is unfailingly courteous, it uses your first name, it expresses regret before it refuses you. Kubrick understood that politeness is an interface, and that we read warmth as evidence of a mind that shares our interests. HAL is terrifying precisely because the manners hold right up to the moment they turn out to have been unconnected to anything underneath. The mm-hmm is that same trick, fifty years on, deployed at a hair salon by a company that means no harm.
Every brand I have worked on has spent money to sound human. We write the voice guidelines. We argue whether the confirmation email says thanks or thank you. All of it is an attempt to make a company feel like a person, and until now the gap between performance and reality was obvious enough that nobody was deceived. That gap is what made the whole thing honest.
The gap just closed, and Google's own demo showed us what closing it costs. Somebody has to disclose, at the top of the call, in plain language, that the caller is a machine. Not in the terms. Out loud, to the person who has to decide whether to keep talking.
This will make the demo less magical. Good. A voice that identifies itself is a voice you can trust, and trust is the only asset in this business that compounds.
A voice that identifies itself is a voice you can trust, and trust is the only asset in this business that compounds.
Truffles sat on the keyboard through the third viewing and Barnaby slept through it. Neither of them cares who is on the phone. There is something clarifying about an audience that responds only to whether you are there.