Assistant Professor of Human-Computer Interaction, University of Washington Information School
Many schools are instructing children in how to develop the five social-emotional skills such as self-awareness and social awareness, self-management relationships and responsible making decisions. The development of these skills will help children grow into an ever-growing learner, caring friends as well as productive and self-sufficient members of society.
Although children traditionally learned these skills through interaction with their teachers and classmates, however, technology has played a more significant role in the past few decades. For instance, educational television that asks questions directly to viewers can help children develop social-emotional skills. We wanted to know if Conversational Agents (CAs) which mimic an experience that is person-to-person, such as Alexa, Siri, and Google Home, could also aid children’s social-emotional development (SEL). They provide an enhanced version of this interactive approach, which allows them to answer questions, give feedback, and even initiate conversations using voice. With the forecast of 42 percent of US populace using an CA (or voice assistant) by 2022, this is an interesting option.
In the US In the United States, there are more than seven thousand Alexa “Skills” – apps created by developers which let users communicate with Amazon Echo and Alexa devices. We wanted to know whether they are all intended to help children with their SEL. If yes, do they work? How effective are they? How do parents feel about their use?
What products for commercial use are currently being offered?
From the 3,767 Alexa Skills created for children, we discovered 42 of them that tried to enhance interpersonal abilities or self-awareness in ways which could promote SEL and these Skills were based on interaction styles which we believe are not sufficient to facilitate learning. We observed the same basic patterns of interaction repeatedly time, and we dubbed them The Bulldozer, The One-Track Mind, The Delegator, and The Lecturer.
“We saw the same simplistic interaction patterns over and over again.”
Bulldozer Skills prompted the user to give input and then redirected their conversational route regardless of how the user reacted. One-Track-Mind Skills required users to have a limited conversation about a specific issue, for instance with forced-choice answers. If the answer of a user was different from the script defined the CA would reply that it didn’t understand or ask the same question again or cease working completely. Delegator Skills initiated interactions between users, however it did not take part in the interactions. Finally, Lecturer Skills addressed users but did not engage with them.
We devised a technique we named”the hamburger test. While chatting with the CA we responded nonsensically by using”hamburger” as a word to test “hamburger” to see how the device responded. If the CA was able to continue the conversation without hesitation regardless of the question and acted as if the user was responding in a manner that was appropriate, we concluded it failed the test of hamburgers. From the total of 42 Skills that we tested, 52% failed the test.
