Speech Technology Magazine

 

Keep Bad ASR Out of My New Dumb Car

Cars equipped with the latest technology, including speech recognition, can complicate your journey
By Moshe Yudkowsky - Posted Feb 15, 2016
Page1 of 1
Bookmark and Share

I need to buy a new car soon, since my current car costs more and more to keep running. This is an important purchase for two reasons: First, I enjoy driving; second, other drivers are trying to kill me.

I drive a mid-engine two-seat convertible with a Lotus engine on a Lotus frame, with a top speed of 55 miles per hour—in second gear. For those of you who don't use a manual transmission—only 7 percent of cars in the United States are sold with them now—this means when you stomp on the accelerator, the car responds. Quickly.

But driving a sports car can get me out of trouble. I recall a recent encounter with a driver whose couch dropped off a trailer. The furniture was sitting in the middle lane—my lane—of a three-lane highway, just as I came around a curve, traveling over 70 miles per hour. I was able to maneuver around it with ease.

Why buy a car now? In part to shop in advance of a need so there won't be time pressure when my present car gives up the ghost. But also because I want to find a nice dumb car.

For example, my car used to have Daytime Running Lights. This "safety" innovation sounds like a good idea to a city slicker—clearly someone who's never almost not hit a deer when it froze in the middle of the road because of oncoming headlights. This feature keeps your car's lights on no matter what. I disabled them on my current car. Will a new car let me?

I also dread the collision avoidance feature: sensors and automated braking systems to avoid obstacles. Would collision avoidance have applied the brakes when I encountered the couch? What about the time a car spun out of control into my lane from two lanes over—could I have accelerated into the next lane and out of the way?

Looking down the road, I will not ride in a self-­driving car for at least five years after commercial release, and never for a month after any major software update. I can just imagine the bug reports: "If you make a left-hand turn onto a street that has a name that begins with a consonant on the last day of month with an odd number of days, then if you make a right-hand turn the next day onto a street with a name that begins with a vowel...".

As for the user interface, I want a dumb one of those, too. My sister's hybrid-electric car has a touchscreen interface that, each and every time she starts the vehicle, demands that she read a warning and agree to the terms. How she manages not to drive her fist through the screen is beyond me. Lawyers dictate the user interface?

Car manufacturers don't want anyone making changes to software, claiming that copyright protects it. At least for now, the right to tinker with software remains; but the law could change, and I could be stuck with an unmaintainable user interface.

Finally we turn to speech recognition. Late in 2014 the American Automobile Association (AAA) published a study of speech recognition systems in various cars. Now, I'd love to have hands-free control over my radio—it would let me keep my eyes on the road and my hands on the wheel. But according to the AAA, not all speech recognition systems are equal, and some introduce rather heavy cognitive load.

The AAA found that built-in navigation systems with perfect automatic speech recognition (ASR) introduced more distraction than simply talking with a passenger. When ASR is imperfect, the system introduces more distraction still, even more than listening to TTS emails and text messages.

A dedicated button for a single function puts limits on the cognitive load; my car audio, with its multitasking buttons, is dangerously more complex. A single-function button also limits the number of tasks to manage, because you run out of room for buttons.

I hate to say it, but unless the ASR is very good, the more complex user interface that comes with it constitutes an attractive nuisance. I don't want to be tempted to listen or compose emails, fiddle with a screen to learn my exact gas mileage, or do any other nifty thing these systems offer. I want to keep my eyes on the road and watch for the next attempt to kill me.


Moshe Yudkowsky, Ph.D., is the president of Disaggregate Consulting and author of The Pebble and the Avalanche: How Taking Things Apart Creates Revolutions. He can be reached at speech@pobox.com.

Page1 of 1