MIT researchers find robots with human-like social behavior earn more trust
A family gathers around their kitchen table to unpack a newly purchased digital assistant. They are more likely to trust a new voice user interface, whether it is a smart speaker such as Amazon's Alexa or a social robot such as Jibo, if it displays human-like social behavior, according to new research conducted by scientists at the MIT Media Lab.
The researchers found that family members tend to view a device as more competent and emotionally appealing if it can display social cues, such as moving to orient its gaze toward the person speaking.
In addition, their study showed that branding, in particular whether the manufacturer's name is associated with the device, has a significant effect on how family members perceive and interact with different voice user interfaces.
The researchers found that when a device has a higher level of social embodiment, such as the ability to give verbal and nonverbal social cues through movement or expression, family members also interact with each other more often, engaging with the device as a group.
Their results could help designers create voice user interfaces that are more engaging and more likely to be used by family members at home, and improve the transparency of these devices. The researchers also outline the ethical issues that may arise from certain personality and embodiment models.
“These devices are a new technology entering the home, and they are still very little studied. Families are at home, so we found it very interesting to look at this from the perspective of generations, including children, grandparents and grandfathers. We were very interested in understanding how people perceive these devices and how families interact with them together”
"The human-centered ideas in this work are relevant to the design of all kinds of personified AI devices, from smart speakers and intelligent agents to personal robots," said Cynthia Breazeal, director of MIT RAISE and director of the Personal Robotics Group.
This work grew out of an earlier study in which scientists examined how people use voice user interfaces at home. At the start of the study, users got acquainted with three devices before taking one home for a month. They noticed that people spent more time interacting with the social robot Jibo than with the Amazon Alexa and Google Home smart speakers. The scientists wondered why people interacted more with the social robot.
To find out, they designed three experiments involving family members interacting as a group with different voice user interfaces. The studies involved 34 families, comprising 92 people aged 4 to 69.
The experiments were designed to mimic a family's first encounter with a voice user interface. Participants were videotaped as they interacted with the three devices, carrying out a list of 24 actions (for example, "ask about the weather" or "try to get an opinion"). They then answered questions about their perceptions of the devices and categorized the voice user interfaces.
In the first experiment, participants interacted with the Jibo robot, Amazon Echo and Google Home without any modifications. Most considered Jibo much more sociable, reliable and responsive. According to Ostrowski, because users perceived Jibo as having a more human-like personality, they were more likely to interact with it.
In the second experiment, the researchers aimed to understand how branding affects participants' perceptions. They changed the "wake word" (the word a user says aloud to activate the device) of the Amazon Echo to "Hey Amazon!" instead of "Hey Alexa!", but kept the wake words for Google Home ("Hey Google!") and the Jibo robot ("Hey Jibo!").
The scientists also provided participants with information about each manufacturer. When branding was taken into account, users considered Google more reliable than Amazon, even though the devices were very similar in design and functionality.
“This also dramatically changed people's view of the Amazon device as competent or as a companion. I did not expect there to be such a big difference between the first and second studies. We did not change the abilities, how they function, or how they respond. The fact that they knew the device was made by Amazon greatly changed their perception”
Changing a device's wake word may have ethical consequences. According to her, a personalized name that can make a device seem more social may mislead users by masking the connection between the device and the company that made it, which now has access to user data.
In the third experiment, the team wanted to see how interpersonal movements affect interaction. For example, the Jibo robot turns its gaze toward the speaker. For this study, they used Jibo along with the Amazon Echo Show (a rectangular screen) with the wake word changed to "Hey computer," and the Amazon Echo Spot (a sphere with a round screen) featuring a rotating flag that sped up when someone said its wake word, "Hey Alexa!"
According to Ostrowski, the modified Amazon Echo Spot did not seem more appealing to users than the Amazon Echo Show, suggesting that repetitive movements without social embodiment may be an ineffective way to increase user engagement.
A deeper analysis of the third study also showed that users interacted with each other more, for example by looking at one another, laughing together or having side conversations, when the device they were using had more social capabilities.
“At home, we were interested in how these systems encourage interaction between users. People are always very concerned about this: how will these devices affect people's relationships? We want to design systems that can foster more thriving relationships between people”
The researchers used their findings to outline several design considerations for voice user interfaces, including the importance of developing warm, sociable and thoughtful personalities, understanding how the wake word affects user acceptance, and conveying nonverbal social cues through movement.
Building on these results, the researchers want to continue studying how families interact with voice user interfaces that have different levels of functionality. For example, they could conduct a study with three different social robots. They also would like to replicate these studies in a real-world setting and explore which design features work best for specific interactions.
