A collaborative Stevens study shows that voice-assistant technologies hold promise for some drivers — and may rub others the wrong way.
Artificial intelligence-based voice assistants for drivers are on the very near horizon.
In fact, the technology to provide automatically voiced warnings about road conditions, offers control adjustments while we drive — or even chat with us when we’re bored or getting drowsy — already exists.
But if we had that voice in our cars, would we actually listen? Stevens professor Sang Won Bae sought to find out.
In a new study co-sponsored by Hyundai and conducted with researchers at South Korea’s Yonsei University, the team examined human drivers’ reactions to three different types of AI voice assistants (known as PVAs, or proactive voice agents) as the volunteers operated a lifelike driving simulator in a lab.
The verdict?
“It depends," says Bae, “on the type of agent that’s talking to you, what it’s trying to do… and the individual driver’s age and experience level.”
Some drivers like it; some don’t
Before recruiting study participants, the team used IBM Watson to help design three types of voice agents.
One, a so-called task agent, asked drivers about climate control, navigation, cabin comfort and offered to do things such as play music, tune in to news radio, or read text messages aloud. A second version, known as a social agent, discussed “small-talk” topics (personal backgrounds, weather, etc.) with drivers as they drove — and even chatted sometimes when the drivers weren’t very talkative. A third version known as the companion agent combined both those task and social functions.
There was also a control agent that could respond to commands, but did not proactively offer any information or conversation to the drivers.
After loading road and highway routes and weather conditions into the simulator, the researchers conducted two studies with human subjects. In the first study, 24 drivers used the simulator; after refining the agents to make them more natural-sounding, seven different drivers operated the simulator a second time with the updated system.
“The first insight we gained was that most of the participants felt uncomfortable with the style of an automatic voice talking to them,” notes Bae. “They actually felt pressured to respond."
"Maybe this is because it’s a new technology and we’re just not used to this yet, or maybe it’s due to psychological factors. In any case, it’s a useful insight for developers.”
Among the three agents, the task agent — the one that simply interacted with drivers to ask them if they wanted adjustments — fared worst, by a large margin.
“People did not get a good driving experience with these task-only agents,” explains Bae.
However, adding small talk to the mix seemed to help… for some drivers, anyway.
“The social agent was fairly well received, as was the companion agent,” Bae continues. “However, when we analyzed the responses of drivers with different experience levels, a sharp difference in comfort became apparent.”
Younger drivers: more pro-PVA than experienced ones
Less experienced drivers — who were also generally younger — responded much more favorably to the offers of assistance and chat than their more experienced driving counterparts.
“This could be due to an age gap and a comfort-with-technology gap,” suggests Bae. “Or it could simply be that experienced and better drivers don’t want or need help, and even see it as an intrusion. This is consistent with previous findings that younger drivers have more trust in technology than older drivers.”
In fact, older and more experienced drivers basically only wanted the PVAs to chat when they became bored or drowsy, to help them stay awake.
"That somewhat suggests they didn’t want the additional information or help operating their vehicles — features that PVAs are fully capable of providing," continues Bae.
This, she says, means “younger yet experienced drivers are expected to be key potential users of PVAs in future vehicles, since these companion PVAs — and their proactive interactions — offer new opportunities for developing personalized intelligent assistants.”
The National Research Foundation of Korea (NRF), funded by the nation’s Ministry of Education, also supported the study, which was reported in the International Journal of Human-Computer Interaction in January.
