Was Google’s I/O unveiling of Duplex, um, duplicitious? The company showed off the technology
, designed to let AI-infused software place phone calls and make appointments, by playing two recorded conversations, to a hair stylist and a restaurant. CEO Sundar Pichai emphasized that they were “real” calls.
As journalists have raised questions about the recordings—for instance, why don’t the establishments identify themselves or ask for contact info?—Google has chosen not to address them. Now Daring Fireball
’s John Gruber has written a post
(one of several he’s posted on the topic
) detailing his suspicions that Duplex might be real, but not yet as refined as the I/O presentation suggested.
I’m not here to ruminate about Duplex, its rollout, and Google’s responsibilities. (For one thing, I did so in my last newsletter
.) Instead, I’m moved to noodle around on a point in Gruber’s post: “A recording is not
He’s right. (I was guilty of lax terminology when I called the Duplex segment of the I/O keynote a “demo” in my initial coverage
, and feel sheepish about it.) An onstage presentation involving playback of audio (or video) is not the same thing as one in which someone uses a product live. It removes the risk of embarassing failure and generally strips away the context that would allow the audience to judge just how impressive a new product or technology really is.
Now, there are legitimate reasons why a company might choose to play a recording rather than perform a live demo. For instance, a true unstaged demo of Duplex onstage at I/O would have been a nonstarter for any number of reasons, starting with the fact that Google wouldn’t have subjected some unsuspecting small business to being part of an onstage presentation streamed to a global audience. And if the business was expecting the call, any interaction with Duplex would have been kabuki rather than a real encounter.
But the thing is, even demos that really are demos are works of stagecraft. Nobody’s going to show you a half-finished feature or delve into functionality that will, more likely than not, result in a crash. They’ll show you precisely what they want to show you, and you generally don’t have a clue what you aren’t seeing.
There’s a Potemkin Village
-like aspect to the entire ritual. That was especially true back in golden age of computer magazines, when companies previewed upcoming products to the press months before they shipped. Almost nothing
was anywhere near ready for prime time, which meant that demos—even ones of genuinely impressive products—were akin to someone driving down a twisty road knowing that even one wrong turn would have sent the car careening off a cliff.
Mostly, companies are able to pull off such feats in catastrophe-free fashion, which is why the major exceptions are so memorable. The 1998 Microsoft scanner demo which blue-screened a Windows 98 PC as Bill Gates watched was big news at the time; 20 years later, the Register was moved to publish an article marking its anniversary.