Image: pop culture geek/Flickr
There’s a moment in Spike Jonze’s Her when Joaquin Phoenix’s character asks his talking operating system why a talking operating system sounds like she’s sighing—in reality, she never breathes, after all. The operating system gets sort of flustered and answers, “That’s how people talk.” Contrast that with 2001: A Space Odyssey—HAL 9000 doesn’t sigh; he doesn’t betray unsureness.* HAL might be sorry, but he’s just not going to open the pod bay doors, Dave. “This conversation can serve no purpose anymore,” he tells Dave. “Goodbye.”
What science fiction reveals, science confirms: Researchers from Carnegie Mellon and Cornell demonstrated that if we’re ever going to get to the point where people regularly listen to their computers’ advice, the robotic voices are going to have to sound less like HAL and more like ScarJo. Robots giving instructions need not sound breathy, but they better not sound too sure of themselves.
The reason for this is related to two facets of human fallibility—our imperfect speech and our fragile egos, which abhor being talked down to by a machine.
When people talk, we, you know, use “discourse markers” that don’t really add much meaning to what we’re saying, but just let other people know that something else is coming up—these are your “welps,” your “uhms,” your superfluous “you knows,” and what not. We almost all also “hedge” our speech, especially when giving advice with “probably,” “I think,” and “maybe,” but not because we’re worried about being wrong.
“People use these strategies even when they know exactly,” Susan Fussell, associate professor of communication at Cornell, said. “It comes off more polite.”
The researchers discovered that those little utterances can make the difference between love and disabling a robot helper. They made videos of an “amateur baker” making cupcakes with a helper giving advice—in some videos the helper was a person, and in some videos the helper was a robot. Sometimes the helpers used hedges or discourse markers, and sometimes they didn’t.
You can probably already guess that the humans and robots that used hedges and discourse markers were described in more positive terms by the subjects who watched the videos. It’s somewhat more surprising to hear that the robots that hedged their speech came out better the humans who did. Fussell attributed this to the charm of a robot saying something unexpectedly casual.
The ultimate message, the researchers said, is that designers of robots that deliver verbal advice should use language that makes the recipients more at ease. As the relationship with our electronics more closely mimics human communication, more cues will be lifted from human interaction and human communications studies.
It’s something that interface developers have been working on already. When Siri doesn’t understand, she says “I didn’t quite get that,” which is much less aggravating than something like “Inadequate vocal reading. Please repeat” or “Speak up, dummy,” although the latter might be sort of charming. Does Siri have a profane setting yet?
Contrast that with the first computer that my family ever had, which used to just really really piss off my mom—she took the error messages personally for some reason, as though the computer was blaming her. There was no malice in that Macintosh LC-475, but the brusque technical language, on top of frustrating technical problem, lead to (sort of hilarious) outbursts of rage.
So, you know, programmers and robots should, well, take note. As the Nixon-Kennedy debates are said to have proved, we love pauses and affectation. People said Nixon sounded robotic, but our future robots would do well to sound like, er, Kennedy.
*HAL doesn’t betray unsureness at first, but (SPOILER for a 46-year-old movie set 13 years ago) when Dave is going to shut him down, HAL says, “I’m afraid,” and it’s utterly terrifying, probably for the same reason that hearing robots speak casually is so delightful.