Siri Co-Creator Questions Texas A&M Texting-and-Driving Study

4/30/13Follow @wroush

The co-inventor of Siri has a bone to pick with the media.

A rash of stories last week—reporting on a study released by the Texas A&M Transportation Institute (TTI) in College Station, TX—implied that using voice-to-text mobile applications such as Siri and Vlingo while you drive is no safer than texting while driving. But Adam Cheyer, the computer scientist who co-invented Siri and sold it to Apple in 2010, says that’s the wrong conclusion to draw from the study.

Here’s what the Texas A&M researchers actually discovered: when drivers on a closed course were holding an iPhone or Android phone in one hand and interacting with Siri or Vlingo, it took them almost twice as long to respond to outside events. For sending and receiving text messages, voice-to-text applications “do not increase driver safety compared to manual texting,” concluded study author Christine Yager, an associate transportation researcher at TTI’s Center for Transportation Safety.

But that’s not how Siri is designed to be used in a car, Cheyer argues.

“I don’t think that there is any evidence that shows that if Siri and other systems are used properly in eyes-free mode, they are ‘just as risky as texting,’” he says.

The crucial phrase in Cheyer’s quote is “eyes-free mode.” When a driver is using Siri with a Bluetooth headset or speaker—as Apple recommends—the app goes into a special mode that limits interactions to voice only. “It assumes you are ‘eyes-busy’ and responds differently,” says Cheyer, who left Apple in 2012.

Vlingo has a similar mode on Android phones. To properly test whether hands-free use of mobile phones while driving is safer than handheld use, the researchers should have asked subjects to use the smartphones in the eyes-free mode, Cheyer says.

“Of course your driving performance is going to be degraded if you’re reading screens and pushing buttons,” he says; that’s why the engineers behind Siri came up with a separate set of voice prompts that don’t require drivers to look at the screen. If more reporters had read the TTI report, Cheyer says, they might have noticed that the voice-to-text software’s capabilities weren’t getting a full test.

Special pleading from someone who still sees Siri as his baby? Maybe, but a source with connections to Burlington, MA-based Nuance, owner of Vlingo, shared similar concerns about the study. (Neither Apple nor Nuance have commented officially on the TTI findings.)

“My goal is not to knock this particular study,” Cheyer says. “I’m just dismayed that the message being communicated by news media—that ‘Siri is just as risky as texting’—is misleading.”

I contacted Christine Yager at TTI last week to ask what she thought about Cheyer’s points.

“We welcome input from the telecommunications industry and look forward to working together to keep travelers safe,” she replied in an e-mail.

But the study was designed to examine the way drivers actually use smartphones in their cars, she said, not the ideal scenarios suggested by software makers or device manufacturers. “We tested the applications in a way that is consistent with how many drivers typically use them,” Yager said.

Many smartphone owners may not even know that Siri, Vlingo, and similar apps have an eyes-free mode, Yager added. “We examined the product information contained in the packaging for the iPhone 4S, and were not able to find information related to the directed mode use of the device,” she said. “The only somewhat relevant reference said, ‘Consider using a compatible hands-free device with iPhone. Use of a hands-free device may be required in some areas.’”

It’s widely accepted that manual texting and driving are a deadly combination. About 18 percent of all vehicle crashes in 2010 involved distracted drivers, according to the Centers for Disease Control and Prevention. Texting is considered one of the worst forms of distraction, since it usually requires drivers to take their eyes off the road, their hands off the wheel, and their minds off the task of driving.

But voice-to-text apps have the potential to eliminate the visual and manual distraction from texting, if not the cognitive distraction. In Siri, sending a text message is as simple as telling the app something like, “Send a message to Mary: I’m going to be late for dinner.” Vlingo works in much the same way.

Yager’s study is thought to be the first to evaluate the safety of voice-to-text apps as an alternative to fully manual texting.

In the study, 43 drivers navigated a 3.8 mile course along the runways of a former Air Force base that now serves as Texas A&M’s Riverside campus. They were driving a 2009 Ford Explorer equipped to record the drivers’ response times, gaze direction, speed, and lane position.

Yager and her team studied four scenarios: a baseline 30-mph drive with no texting, a drive that included manual texting tasks, a drive using Siri on an iPhone 4S, and a drive using Vlingo on an Android phone. A green LED light was mounted on the Explorer’s dashboard, and drivers were instructed to press a button every time they saw it come on. Their response times served as a measure of their distraction levels.

For the voice-to-text tasks, drivers were using the phones just as they would if they’d been standing on the sidewalk: that is, by holding the smartphone in one hand, activating Siri or Vlingo with a manual button press, speaking a message, proofreading it visually, correcting it if necessary, and sending it with another button press.

It would have been reasonable to expect to find slower response times among the people who were sending texts the old-fashioned way. And in fact, that’s what Yager’s team found—-their responses were delayed by a factor of 1.92.

The more surprising finding was that using Siri and Vlingo didn’t improve the situation much. “When texting using Siri, response times were delayed by a factor of 1.87; and when texting with Vlingo, response times were delayed by a factor of 1.77,” Yager’s report states.

Drivers also tended to slow down when they were texting, and to spend less time watching the forward roadway. The effects were equally severe whether the drivers were texting manually or using the voice-to-text apps.

In the end, Yager concluded that the voice-to-text apps are no less impairing to drivers than manual-entry texting.

And that’s where Cheyer’s objection begins. It’s already well known that reading small text and manipulating virtual buttons on a screen “decreases attention on the road and increases driver distraction-related incidents,” Cheyer says. That’s why 10 states have banned handheld use of mobile phones while driving.

But in eyes-free mode, visual and manual tasks are reduced to a minimum. “The study seems to have misunderstood how Siri was designed to be used,” Cheyer says.

In regular mode—the way Yager’s subjects were using the iPhone—Siri shows a text message on screen for proofreading and confirmation. But when used with a microphone—either on a Bluetooth headset or on the earbud cord that comes with the iPhone—Siri reads back the message using text-to-speech technology.

In car mode, Cheyer says, “Siri is even stricter and will not execute certain commands that would require looking at a screen.” Vlingo also has a hands-free mode in which all text messages and prompts are read aloud using text-to-speech.

It’s possible that this kind of hands-free, eyes-free texting still causes some level of distraction—but that’s not what the TTI study tested, and to find out, more research would have to be done, Cheyer says.

The TTI study shows only that “non-eyes-free voice input is not a significant win over non-eyes-free manual texting,” he says. “This is the message that should be transmitted to people.”

Perhaps the biggest point in defense of Yager’s study design is that it was realistic. Most people who text while driving—whether or not they use a voice-to-text app—are probably holding and looking at their phones while they do it.

But Yager says her team is definitely interested in studying how voice-to-text applications involving a Bluetooth speaker or headset might affect driver behavior. “Understanding the distracted driving issue is an evolving process, and this study is but one step in that process,” she says.

Wade Roush is a contributing editor at Xconomy. Follow @wroush

By posting a comment, you agree to our terms and conditions.

  • Rich Hoyt

    Just how distracting was watching for the green LED and pushing the button? That needed to be tested speratly, the subtracted from the effect of “Siri” texting. The measure of reaction time is not realistic compared with normal driving. They should have measured normal like the reaction to stop lights for example. Also – “18 percent of crashes involved distracted drivers”, but what percent of the 18 were because of texting – there’s a whole lot of other distractions like eating or putting on makeup. Quit overstating the risk – texting is low compared with the overall causes of accidents.

  • Paull

    I run a limo type service, and a regular client is blind since birth. he and Siri are good friends, ALL his phone usage is eyes free. Needs to be seen and heard to be fully appreciated