After his initial arrest, Joe is led by a noose stick (and a police officer speaking some devolved version of copspeak) to a machine to get an identity tattoo. Joe sits in the chair and a synthesized voice says, “Welcome to the Identity Processing Program of America. Please insert your forearm into the forearm receptacle.” Joe does as instructed and it locks his arm into place. A screen in front of him shows the legend “Identity Processing Program of America” superimposed over an USA pattern made up of company names and Carls Junior amputated star logos. Five rectangles across the top are labeled: System, Identity, Verify, Imprint, and Done.
It prompts him to “…speak your name as it appears on your current federal identity card, document number G24L8.” Joe says, “I’m not sure if—“ The machine interprets this as input and blinks the name as it says, “You have entered the name ‘Not Sure.’ Is this correct, Not Sure?”
Joe tries to correct it, saying, “No…it’s not correct.” On the word “correct” it dings and continues, “Thank you. ‘Not’ is correct.” “Not” stops blinking in the interface.
“…Is ‘Sure’ correct?” Joe has patience and tries to correct it. “No, it’s not. My name is Joe.” It blurts out some error beeps. “You’ve already confirmed your first name is Not. Please confirm your last name, ‘Sure.’”
Joe: My…My last name is not Sure.
Kiosk: Thank you, Not Sure.
Joe: No. What I mean is my name is Joe.
Kiosk: Confirmation is complete. Please wait while I tattoo your new identity on your arm.
The machine begins to shake and make noises. Joe says, “Wait a second. Can we start over? Can I cancel this?” He sees a progress bar, labeled “Tattoo In Progress…”
“Can we cancel this and just go back to the beginning? They’re gonna tat—Ow. Could I speak to your supervisor? Ow!” While he’s trying to wrench his arm free, the machine instructs him, “Please hold still for your photograph.” It flashes an unflattering picture of him and the clamp on his arm releases. He removes his arm to see the new tattoo. The screen shows him his identity card.
In exasperation from the whole ordeal he mumbles, “Oh, that’s fuckin’ great.”
This scene is played for the Vaudevillian yuks, but it does illustrate some problems with conversational design. And note that, if you’re interested in this topic, let me make an early shout out to the book Conversational Design by Erika Hall, published earlier this year.
When it hears Joe say “I’m not sure…” it takes it as a literal answer. It does not recognize that Joe isn’t answering the question. It is one meta-level up. He has a question about the question being asked. Humans are pretty good at recognizing when another human is breaking the usual logic of adjacency pairs and not providing an answer to the question. (This was discussed in Make It So in Chapter 5, “Gestural Interfaces” in relation to Minority Report) Computers have a harder time of it. If this kiosk understood it, it would be know that he’s not answering the question, and resolve what conversational analysis calls “the expansion” before returning to the question. (Disclosure: I work there and know the guy who wrote that.)
Aside: Douglas Hofstadter in his mind-expanding book Godel, Escher, and Bach, writes about the trick question “Have you stopped beating your spouse?” for which neither “yes” nor “no” are good answers, but the only “correct” answers according to the binary frame of the question. In that text he introduces the eastern answer of “mu” (or “wu” in Chinese languages) that means roughly “the answer does not fit the question.” So it can be said that computers have a hard time understanding mu.
Designers of digital assistants have to wrangle with this, but it’s rarely a problem that the individual designer must wrestle with. Language and naming are informal, slippery notions as far as computers are concerned, so it’s understandably a hard problem. It’s entirely possible that someone has chosen “Not Sure” as a name, but it’s highly unlikely. And that’s another problem.
Understanding intent might be a little easier if the computer could recognize that “Not” and “Sure” are unlikely values for a name. (Even in Idiocracy where names tend to be brands like Lexus, and Frito, and Biggiez. More on this later.) If it knew that, it would have a low confidence that it “heard” correctly, and shift into a repair or at least clarification mode. “‘Not’ would be very uncommon name. Let me be extra careful, here…” It could even shift into a more deliberate mode of input, like a keyboard, or asking him to spell his name out (or, you know, cancel the whole thing.)
When the kiosk is asking for confirmation, it hears Joe say, “No it’s not correct” and registers the keyword “correct” but misses the function word “not” which completely flips the meaning.
Again, avoiding this speech-to-text error would be a developer’s task, but dealing with the back-and-forth would definitely fall to a designer. When clarifying low-confidence input, users should be able to provide discrete high-confidence feedback.
Joe could, for instance, be shown the kiosk’s (stupid) understanding of his input and—since this has a pretty permanent consequence—wait on his confirmation and providing the simple option to redo it so he can try some other tactic to getting “Joe Bauers” in there until he gets it right.
But, of course, this is Idiocracy, and Joe is stuck with it.
But we’re not
I mean, Republicans have done all they can to suppress votes that don’t favor them. They don’t care about Democracy or the will of the American people as they do staying in power to serve their 1% overlords. But research shows that people who have a plan to vote are more likely to actually do it, and if we all do it, we can overwhelm them with sheer numbers.
There are lots of tools to help you make a plan, but let’s send some traffic to our friends at Planned Parenthood. They’ve been under a lot of pressure during this administration. Maybe throw them a few sheckles while you’re there. Not for the election, but because you’re a good person. And vote all of them out.
Pingback: Frito’s F’n Car interface | Sci-fi interfaces