How Can We Make Conversational Agents More Humanlike?
Research shows reflexive behaviors make robots more lifelike and appealing.
Posted Jan 14, 2018
Conversational agents that can chat with you, such as Alexa from Amazon, are now commonplace. Robots, too, can have conversational abilities — for example, Pepper by Softbank and Jibo by Jibo Inc. Robotics researchers try to make communication between robots and humans smoother by basing robot behaviors on human behaviors. For example, many studies have based robots’ eye gaze and head tilt behaviors on human models as ways to communicate interest and intent. Fewer studies have investigated the effects of robots' expression of other reflexive (automatic and involuntary) behaviors on human responses.
A recent study by researchers from Ewha Womans University in Seoul and the Korean Institute of Robot and Convergence investigated whether a robot that communicated its internal and external states via reflexive behaviors would be seen as more humanlike, and as providing better service.
In the first study, participants were shown three robots in a dusty environment. The first robot commented that there was dust in the air, the second robot coughed, and the third robot did not react in any way. Participants rated the coughing robot as more humanlike and lifelike, and they were more satisfied with this robot than with the other two robots.
In the second study, participants were shown three robots that were running out of batteries. The first robot commented that it was out of batteries, the second robot made a stomach grumbling noise, and the third robot did not react at all. Again, the robot that made the reflexive body noise was seen as more humanlike and more lifelike, and people were more satisfied with it than with the other two robots. The effects were slightly larger for the tummy grumble (reflecting an internal state) than for the cough (reflecting the external environment).
This study suggests that people perceive robots that express bodily symptoms in these situations as behaving more naturally than if they verbalize their states or do not communicate at all. This makes sense, because in a dusty environment people could already see the dust for themselves and they would not need to be told about it. Coughing is a natural reaction that suggests the robot is affected by the dust as a human would be. Likewise, when a person is hungry, they will not always say so, but their stomach will often give them away.
What does this mean for our future digital companions? Does this mean we want robots that burp, hiccup, and pass wind? Or are these behaviors more susceptible to the norms of politeness? Watch this space...
Kang, D., Kim, M-G., & Kwak, S.S. (2017) The effects of the robot’s information delivery types on users’ perception toward the robot. Presented at the 26th IEEE International Symposium onRobot and Human Interactive Communication (RO-MAN) Lisbon, Portugal, Aug 28 - Sept 1, 2017. pp 1267-1272.