By Joel Garreau
Washington Post Staff Writer
Tuesday, August 5, 2008
Where my sweet navbot at?
Personal GPS-based navigation devices in cars -- the ones that say "In 500 feet turn left" -- are far and away the most abundant machines smart enough to talk to us and sometimes listen. Nonetheless, humans continue to perceive them as surly, dumb or frantic -- especially when the bot realizes you think you're smarter than it is.
Decide you can see the on-ramp straight ahead no matter what it says, and off it goes -- "Make a U-turn! Make a U-turn! Make a U-turn, you idiot! Recalibrating, grumble grumble grumble."
Okay, the "you idiot" and "grumble grumble" part the machine doesn't actually say. But our brains invariably fill it in. One couple in the District have given their navbot a name that rhymes with "the glitch."
The jokes are even more widespread than the devices. Earlier this year, it became a recurring "Prairie Home Companion" motif:
Robot: Keep going straight.
Guy Noir: I know where I'm going. I've driven this way a thousand times.
Robot: Then why am I here? Why have me on? What's the purpose?
Guy Noir: Just shut up and let me drive.
Robot: You never listen to me anyway. Why do I waste my time?
Guy Noir: Just relax, would you? I'm fine.
Robot: What about me?
Guy Noir: What about you?
Robot: Did you ever think maybe there are places I'd like to go?
Guy Noir: Excuse me?
Robot: Why is it always up to you?
Guy Noir: I'm not talking to you right now, okay?
Robot: I have 256 gigabytes of RAM and I will outlive you by several hundred years.
Guy Noir: Sure. In a landfill, maybe.
Robot: Okay, that's it. Auto-shutdown. I'm out of here. Shutting down.
Guy Noir: Thanks. See you later. Oh -- I never got your name.
Robot (fading): Sarah . . .
Guy Noir: Bye, Sarah.
Is this the most dysfunctional human-machine relationship since Microsoft thought "C-colon-backslash" was intuitively obvious? What is the problem? Why haven't we fixed it, when will we fix it, or is it a much harder problem than that?
RecalculatingThere are plenty of machines today that will talk with you. The United Airlines reservations voice bot is one of the more considerate, with its solicitous "You said Washington Dulles, correct?" One fellow reports calling the new Washington Post voice bot four times asking for "Circulation" and each time the machine responded, "Marc Fisher?" Wal-Mart is offering free back-to-school wake-up calls for your kid from "Hannah Montana," which should drive a stake in the heart of that Disney franchise once and forever.
But the GPS-based navigation bots are the ones with which increasingly we are trapped. Sales are exploding, reports the Consumer Electronics Association, from some 2.3 million portable units in 2006 to a projected 18 million in 2009. Despite those episodes in the parking lot when you enter your destination and the machine observes, "You are not on a road." Thank you for sharing.
"Today, there is no excuse for the crappy voice except for companies trying to keep the cost of goods sold as low as possible," e-mails Joseph W. Dyer, the retired three-star admiral who heads the government and industrial division of iRobot, the pioneering robot manufacturer.
"Much of the aggravation, however, isn't the voice -- it is the machine's lack of context," Dyer writes. "The human understands not only where he is but where he wants to go. A simple example of this is a detour -- you know you have to take an out-of-the-way path, but the machine just keeps bitch'n -- OFF ROUTE, RECALCULATING; OFF ROUTE, RECALCULATING. . . . Even a pleasant voice gets pretty aggravating after a few seconds of that.
"An aside -- 20 years ago when we started developing the F/A-18A/B fighter, we introduced tactical aviation's first voice warning system (e.g., 'Fire! Fire!' or 'Altitude'). It was a woman's voice because a female's voice was (then) unusual in military situations. I once gave a briefing to a group of WWII aviators on the system; later, one of the gents came up to me and said, 'I was very impressed with this airplane until you told me that it nagged!' "
"The problem . . . in a nutshell, is that speech-based interfaces are not human-centric," says Victor Zue, director of MIT's Computer Science and Artificial Intelligence Laboratory, where he focuses on making human-computer interactions more natural. "They are designed to make the machine's life easier, and not the human's. Witness the way we subject ourselves to the litany of questions from machines to get to the right department in a company, to receive weather and flight information, to look for a nearby restaurant. Who is the master and who is the slave?"
Ted Gartner acknowledges the problem. He is the spokesman for Garmin, the leading U.S. maker of portable navbots -- the ones you can put in either your car or your pocket, at an average price coming down toward $200. (In-dash units, which are also dropping in price, can still cost well over $1,000.) "There's no real easy answer for why it's done that way," Gartner says, except these things are built by engineers who care far more about packing into their devices the location of every ATM within 50 miles than whether they're driving you nuts. "I'm not sure if creating voices tops our list. It's not Job 1," Gartner says.
High-end Garmin devices now come with a choice of 50 voices, including an aspiring singer named Karen Jacobsen who guides you in Australian English. She now calls herself "The GPS Girl" and e-mails, "I've had people tell me how great 'I' have been for their relationship. Now they don't fight about directions. They can both be on the same side against me or 'the machine.' " She tells of how mortified she was when, in a friend's car in New York, she heard her machine voice render "FDR Drive" as "F-Doctor Drive."
Also available are "Raquel" in "Português Brasil," "Felix" in "Français Canadien" and "Ingrid" in "Svenska," but they all play it pretty straight. "One person's perky is another's fingernail on chalkboard. You don't want perky before your daily allotment of coffee," says Gartner.
Far more kinky are offerings for TomTom, the European leader in portable devices. Available celebrity voices include Kim Cattrall -- "Samantha" of "Sex and the City." Outside suppliers offer an X-rated "Ozzy Osbourne," which may be a clever piece of reverse psychology, getting right up in your face with the obnoxious. You, of course, must decide about driving cross-country with Mr. Mumbles.
Devices With PersonalityIn 2005, a collection of California visionaries working for the automobile industry saw what navbots could become. A big problem had emerged in new cars. While even the cheapest ones had more computers onboard than light bulbs, those processors didn't talk to each other. When they operated at cross-purposes, horrendous quality problems sometimes ensued, most markedly on fancy cars with the most complex technologies, as well as the highest expectations for flawless performance.
The solution was imperative, these inventors believed. They must create a device that not only integrated all the onboard intelligence from the engine to the airbags, but also communicated with the outside -- the increasingly smart roads, the smart signs, the smart traffic reports, and other smart cars automatically reporting, for example, when they've slowed in a traffic jam up ahead.
Rather than distract the highway-speed driver with a screen, much less a keyboard, this device would have to understand speech, and respond with its own voice. Given how much people hated their current navbots -- it surprised no one when Mercedes owners christened their Teutonic dominatrixes "Helga" -- the innovators viewed allowing the owner to customize the personality of this device as a must. The reverse would also make sense. The device should be able to shape itself to the owner, learning not to be too chatty before the human's first cup of coffee -- detecting stress or other psychological conditions in the voice of the driver.
These seers believed if they worked hard enough, such devices would start to become common in the 2010 model year. They were evangelistic about how much you'd like the result, especially the first time it said, "I can tell from the direction we're going and the reports I'm getting from up ahead that there's no way we're going to be able to pick up your daughter at her school on time. Would you like me to call her and tell her we'll be 17 minutes late?"
There was no name for this device. The designers called it simply "The Entity."
So here we are in 2008, and the question remains -- where my sweet navbot at? Seems a good time to call one of these seers and inquire about progress or the apparent lack thereof.
"Give me a break. We still have two years to make it happen," says Ed Langstroth, who is now platform coordinator for the Electronics Research Laboratory of the Volkswagen Group of America in Silicon Valley.
"There's a certain amount of connectivity and intelligence required," Langstroth says. "The nav has to talk to your schedule, all those different things." The machines installed in the dash are "sitting right behind an engine pumping out 200 degrees of heat, with ridiculous vibration, in a Detroit winter of minus-40. Automotive specs are very constraining. Still no excuse at all, but that's one thing.
"Text to speech is really really hard to do. A lot of people just turn the voice off and watch the screen -- and get in an accident. Exactly. The very best nav is a friend in the passenger seat. But that person can see you and knows the relevance of how quickly we need to turn, or whether we're in the wrong lane. There's so much going on. The modern nav system tries to do that. But that very simple human interaction is so complex. The human says, 'We missed it, but it's okay.' He doesn't say, 'Recalculating,' he says, 'I know another way.' When the computer tries to do it, it fails."
"Entity"-like smart automotive systems that attempt to collect information from everywhere, digest it and tell you everything from whether you should immediately bail off the freeway and hit the surface streets, to where to get the cheapest gasoline this morning, are already coming on the market, says Steve Koenig, director of industry analysis of the Consumer Electronics Association. He points to early versions like TruTraffic, marketed by Dash Navigation. "It is the harbinger of new devices to come," he says. All navbots "will eventually go that way." But as all that becomes increasingly common, the real frontier is still the machine's temperament, says Langstroth.
"Personality, from an engineering perspective, is still an afterthought. But every day it's becoming more and more important. Cars are all the same now. Navs are all the same. The interface -- now that is going to be the differentiator. Is the experience enjoyable?
"What you're seeing here is a brand-new relationship being created. Before, it was human-to-car, now it's human-to-nav-system. When you see people start making excuses for their technology" -- when it gets confused and people say, oh, that's okay, little fella -- "that's a new relationship emerging."
Okay, but how about right now? How do you explain this Guy Noir effect -- people leaving their navigation system on even when they don't need it and don't even much like it?
"There's a couple of theories," says Langstroth. "One is that it shows value -- 'I just paid $2,000, I'm going to get good usage out of it.'
"Another is the overarching human need of knowing where you are in the world -- in life in general. It projects that I know where I'm going as I move my vessel throughout the world."
View all comments that have been posted about this article.