Children turn to online search tools to complete classroom-related  assignments. Unfortunately, these tools are rarely explicitly tailored to meet children’s needs and skills. We seek to design a Search Agent
(SA) for children in the context of education; one that can foster natural, effective and rewarding interactions in the educational context. As a starting point, we turn to experts in this topic: children. From them we learn which ones are the pragmatic and hedonistic qualities expected of such SA.

Study sessions

We conducted an user study in two sessions:

• S1: We explored the degree to which experience and familiarity with virtual assistants for search can influence children’s perceptions and expectations of an ideal search companion for the classroom.
• S2: We examined the degree to which experience with educational technology affected children’s expectations on favored traits and behavior for an ideal search companion for the classroom.

In both sessions we asked children to draw their imagined SA; we also asked them fill out a survey to capture their preferences and expectations on pragmatic and behavioral traits of their ideal SA. This survey is based on well-known and accepted practices for data collection related to user preference, but adapted to children’s attention span and skills for completing such a survey.

Results and Findings

From surveys to children in S1 (n = 20) and S2 (n = 35)  the following findings emerge:

• Children in S1 depict their ideal SA as animals or computers. Instead, children in S2 more noticeably prefer an ideal SA to resemblance technology-related devices.
• The majority of the children in S1 incorporate phrases or icons to their drawings in order reflect that they expect an ideal SA to be willing to help and be friendly.
• Children in S1 tend to use feminine traits to depict their ideal SA (such as big eyes with long eyelashes or cute, large bows).
• Common terms  children in S1 use to describe an ideal SA include: friend, homework, and help. For S2 the most  frequent term used on children’s descriptions of ideal SA is talk.
• For children in S1, desired features include SA’s ability to (i) help users have fun, (ii) take care of users’ privacy concerns, and (iii) not distract users — not statistical difference on desired features across genders.
• For children in S2, desired features include SA’s ability to (i) remember previous requests, (ii) take care of users’ privacy concerns, and (iii) make users feel safe — fun also emerged as a favored trait.

Outcomes and Lessons Learned

Desirable features. Children in S1 and S2 fully embraced personification for the SA, yet no clear metaphor emerged from the drawings, descriptions, and surveys in terms of what or who an ideal SA should embody. Children do require for an ideal SA for the classroom to be a friendly-looking, personified agent, one that would be fun to work with, but would not cause too much distraction, and also one that would protect their privacy, while being easy to communicate with.

Differences in depiction of an ideal SA across S1 (experience with search assistant) and S2 (experience with educational technology) align with existing reports regarding the fact that users’ knowledge of engineering influence how an agent is perceived. Further, from their descriptions, we saw that children in S1 expected the ideal SA to be able to help them solve any of their tasks, whereas children in S2 had more realistic expectations—they simply wanted the SA to have good quality communication, comparable to that of other technologies they have encountered in the past.  In the end, age, the level of experience, familiarity, and technical expertise are determinant factors when outlining the behavior of an ideal SA (which was also observed in existing research studying families).

Inspired by existing protocols for children and adults, along with lessons learned emerging from these works and our prior experience, we designed a study protocol tailored for children. We carefully outlined dimensions to elicit desires and expectations concerning age and experience in the classroom context, while enabling direct comparisons across experiments. This protocol  can be adopted to explore other aspects that can lead to differences in desired traits for SA.

Acknowledgements

We appreciate the participation in our study of the students in primary schools in Italy and Switzerland.