To explore our research questions, we conducted a series of participatory design workshops. Cultural-responsiveness is not only an analytical frame for our work, but also a methodological one. Accordingly, we leveraged practices from the HCI literature on participatory design and design with children [
15,
22,
29,
72] that allow learners to express themselves in a variety of ways [
63].
3.1 Participants
Participants were 15 girls, recruited from two out-of-school programs, GirlsTeam and RoboGirls (names anonymized for the purpose of publication). GirlsTeam is a national organization that aims to support girls in building community, while discovering new skills across a variety of interest areas, such as science and art. Participants from GirlsTeam were recruited from the science, technology, engineering, and math (STEM) arm of the regional GirlsTeam affiliate. Participants from GirlsTeam had prior interaction with the researchers and each other in a virtual, three-day introductory programming camp [
65,
67]. In this prior camp, participants learned introductory programming concepts while manipulating a virtual robot with blocks-based programming. This prior camp was culturally-responsive, and encouraged learners to reflect on the ways their identity interact with computing. For example, in one activity, learners were asked to program a robot to greet them according to their culture. This coding task was accompanied with conversations about the ways greetings are specific to our identities.
RoboGirls is a city-wide program, focused on empowering girls in STEM, with a specific focus in robotics. RoboGirls is affiliated with a local, private university, with multiple participants having some connection (social or familial) to the university. RoboGirls members frequently participate in local and national robotics competitions. Participants from RoboGirls had no prior interaction with the researchers. However, many had prior friendships with other RoboGirls members. Participants were primarily in middle school (7th and 8th grades), with one 9th grade participant. The study had ethics approval by an IRB and all participants were monetarily compensated for their time. Demographic details about participants with anonymized pseudonyms are shown in Table
1.
3.2 Participatory Design Workshop
To answer our research questions, we conducted a series of participatory design sessions. A participatory design process enlists target learners at all stages of the design process, in order to ensure their perspectives, identities, and values are embedded into the technology design [
72]. We specifically organized design sessions around creative activities that allow learners to ideate and communicate design ideas related to the target questions [
72].
Learners were split into two cohorts, according to the organization they were recruited from (e.g., one GirlsTeam cohort and one RoboGirls cohort). We conducted five participatory design sessions per cohort (10 total), lasting 60 to 75 minutes each. Sessions were scheduled according to learner availability, and took place approximately every two weeks. Learners were encouraged to attended all five sessions, but could attend as they were available. Sessions took place virtually, over Zoom (zoom.us). Learners were encouraged to interact with facilitators and each other via whatever modality was comfortable for them, such as speaking out loud or using Zoom’s built-in chat functionality.
Each session began with a warm up activity, where learners responded to prompts that foreshadowed the day’s activities. Each session then had a focal creative activity. For these activities, learners were primed to think about creating a robot companion. Activities were grounded in this idea, as learners designed different aspects of their robot companion (e.g., dialogue, appearance, tasks). Finally, all sessions ended with reflective discussion questions, themed around the topics explored that day. Each session’s content was independent of the previous sessions, to accommodate learners who could not attend a session. However, content in early sessions informed that of later sessions. To facilitate this, the researchers debriefed after each session and discussed what important themes might need to be followed up in the future. Google Slides (slides.google.com) and Google Jamboard (jamboard.google.com) were used as a tool to facilitate design activities. Examples of artefacts generated from sessions are shown in Figure
1.
In the first session, learners were asked to create a two collages [
66] using Google Slides - one that reflected their own identity, and one for that of their robot companion. Discussion questions focused on how learners conceptualized identity, and important aspects of their robot companion’s identity. Both the collage activity and the discussion questions positioned learners as creators of the robot who had full control over robot design.
In the second and third sessions, learners designed conversations with their robot companion. In the second session, pairs engaged in a role play [
64,
69], with one learner acting as themself, and the other as a robot companion. Learners were told to act out a text-based a conversation with their partner, according to their role and a given scenario. For example, learners were asked to have a conversation with their robot companion, in the scenario that the robot did not want to run the code they just developed. learners completed multiple scenarios, and switched which role they occupied. Discussion questions were themed around learner’s reactions to the role play, and other scenarios they might encounter with a robot companion. In these activities, learners were positioned as co-creators as their robot was role played by a peer who exhibited agency.
The third session was an exploration of learner customization of a robot’s language. Learners interacted with different prototypes [
30] for how they might teach their robot companion how to speak with them. In particular, learners were shown two prototypes in Google Slides. In the first, they could send a postcard to their robot companion, with instructions on how they’d like their robot to communicate with them in particular scenarios. In the second prototype, learners engaged in a text message conversation with their robot companion on similar topics. As an example, learners could tell the robot how the robot should respond if it was time to celebrate. Learners put responses such as, "good job, your hard work is paying off" or "wow you are really getting better." Discussion questions were themed around learner motivations for customizing robot dialogue, and interactions that could facilitate that. In this activity and discussion, learners had full control of what they wanted the robot to say, thus they were positioned as creators.
In the fourth session, learners critiqued three prototypes [
15] for how they might customize their robot. These prototypes were created based on learners’ discussions in earlier sessions. The first prototype took a bags-of-stuff [
75] approach, where learners were given a variety of parts with which to create their robot. Example parts included arms, legs, eyes, or shapes with different textures. Learners constructed their robot by dragging-and-dropping parts onto a canvas. In the second prototype, learners created non-physical aspects of their robot’s identity, by typing their preferences into a textbox. For example, learners were asked to add their robot’s interests. In the third prototype, learners only selected one element of the robot - robot eyes - and were shown a variety of options to drag-and-drop onto an existing robot. Prototypes were intended to be mid-fidelity, giving learners a sense of how they might customize the robot in different ways, without building out an entire interactive system. Thus all prototypes were created with Google Jamboard, and interactions were limited to adding text or images to the screen. After interacting with all three prototypes, learners used sticky notes [
15] to indicate aspects of the prototypes they liked and areas for improvement. Discussion questions further probed how learners would react to a robot that expresses an opinion on its customization, as well as understanding learners’ process for making decisions. Learners had complete agency in making decisions about their robot in the activity, thus they were positoned as creators. However, learners were positioned as both creators and co-creators in the discussion questions, as they reflected on their own decision making process and how they’d react to the robot expressing an opinion.
In the fifth session, learners were tasked with creating stories [
23] for how they want to interact with their robot companion. They created stories in response to a series of prompts that we selected based on themes from the earlier sessions and scenarios that encouraged learners to think about where their robot might have a sense of agency. For example, one story prompt was: You created your robot and they think you left off some important body parts. Write a story about what parts the robot wants you to add and why. Discussion questions further probed who can create different aspects of the robot, and how to navigate disagreements. Both the activity and discussion questions positioned learners as co-creators.
3.4 Data Collection and Analysis
We gathered two main types of data to facilitate answering our research questions. First, we utilized all artefacts generated by learners in the sessions. For example, we analyzed stories they created or their robot designs. Second, we analyzed learner responses to the reflective discussion questions posed by the researchers. These responses were recorded using Zoom’s built-in functionality. They were then automatically transcribed through Rev (rev.com), and chat responses interleaved. These automated transcripts were fine-tuned by the first author.
To identify themes related to our research questions, we conducted a thematic analysis of the data [
24,
40,
68]. Two authors discussed initial themes during the inductive open coding process and reached an agreement on initial codes and definitions. Our initial open coding phase identified learners’ opposing views regarding robots’ agency in response to different activities and discussion prompts. Accordingly, authors explored learners’ embodiment of roles as creator or co-creator in relation to different activities in which robots exhibited no agency or some degree of agency. They then clustered initial codes into abstract concepts and organized into categories including the design power of robots and human, the triangulation of appearances, personalities and functionality, as well as the social relationship between human and robot technology. Through comparing the opposing attitudes and tensions under each category while positioning learners as creators and co-creators, multiple authors discussed and confirmed the data supporting themes. These themes were presented to all authors, who collectively fine-tuned and finalized the findings reported below.