In this paper, we present SCAMP - Social Configuration Affordances for Museum Play - an analytical framework we develop and use to highlight the relationship between designed affordances at interactive museum exhibits and different social playful behaviors they trigger and support. We do this through a selective case study analysis of Rainbow Agents, an interactive museum exhibit designed to support play across multiple social configurations. This variety of configurations is valuable for museum settings, as it helps museum visitors engage with each other according to their preferences and also enables the emergence of modes of collaboration and competition novel for learners in such contexts. Our SCAMP analysis of Rainbow Agents sheds light on design features which successfully support different forms of productive social play - including short and long episodes of competitive, collaborative and parallel play, spanning play, teaching, and receiving interpersonal interactions. In particular, we pay attention to behaviors representing a variety of mentoring and learning opportunities - in line with and extending the vision and goals of educational games' and science museum exhibits' designers and researchers.

Complex systems simulations can support collaborative water planning by allowing stakeholders to jointly see hidden effects of land- and water-use decisions on groundwater flow. We adopted a participatory modeling progression where stakeholders learned to modify and use increasingly sophisticated models to assess policy impacts on groundwater levels. Stakeholders' shared understanding of the problem and the novelty, concreteness, and richness of proposed solutions evolved alongside the models' degree of realism, but up to a certain point. More realistic models became a distraction and stymied efforts to plan for water shortages. The reflective learning required to plan for complex environmental problems is best supported by models that strike a balance between representational fidelity and end-user intelligibility. Complicated models and high-resolution data may overwhelm model users, preventing them from acting on the useful planning insights they derived from the exploratory modeling, particularly within social contexts that exhibit strong power dynamics and favor prediction.

Large-group (n > 8) co-located collaboration has not been adequately studied because it demands different conceptual framings than those used to study small-group collaboration, while also posing pragmatic constraints on data collection. Working within these pragmatic constraints, we use video data to devise an indicator of the current possibilities for learner collaboration during large-group co-located interactions. We borrow conceptualizations from proxemics and social network analysis to construct collaborative opportunity networks, allowing us to define the concept of collaborative opportunity temperature (COT) readings: a “snapshot” of the current configuration of the different social subgroup structures within a large group, indicating emergent opportunities for collaboration (via talk or shared action) due to proximity. Using a case study of two groups of people (n = 11, n = 12) who interacted with a multi-user museum exhibit, we outline the processes of deriving COT. We show how to quickly detect differences in subgroup configurations that may result from educational interventions and how COT can triangulate with and complement other forms of data (audio transcripts and activity logs) during lengthier analyses. We also outline how COT readings can be used to supply formative feedback on social engagement to learners and be adapted to other learning environments.

In this paper we define the concept of collective usability, a complex systems perspective on usability that positions an entire group, not an individual, as the unit of analysis. Shared XR experiences have inherent temporal and spatial properties that produce emergent, collective impacts which can impede learners’ engagement. Assembling large groups of users to test multiple design configurations is both logistically and financially impractical, however. We demonstrate the practical value of exploring the design space of an XR experience with a simple observation-informed Agent-Based Model. We used the model to explore how changes in the number of simultaneous users, and in the size, placement, and interaction duration of the proffered interactives, could affect collective access to a large-scale, mixed-reality, multi-user museum exhibit. (Collective access, an element of collective usability, is the degree to which users can gain access to each of the different interactives.) With this simple model, we explored (1) how the bottom-up propagation of individual-level design properties can affect collective outcomes, as when certain interactives’ linger times cause a bottleneck, and (2) how the top-down propagation of collective design constraints can be used to guide individual-level design, as when we determined thresholds for the “stickiness” and “repeat allure” of an interactive to improve collective access. The final design of the exhibit implemented many of the design guidelines uncovered by the model. We argue that collective usability models could be useful for addressing a range of collective usability issues, beyond collective access, for temporally and spatially sensitive XR learning environments.

Engaging learners with complex unfamiliar datasets is a known challenge in Data Science education. One promising phenomenon investigated in related work is perspective-taking. A first-person "actor" perspective can help facilitate group and individual sensemaking by mediating observations and actions taken by learners. Here we investigate how museum visitors made use of an actor perspective when exploring an open-ended, interactive data map museum exhibit. We use a mix of qualitative and quantitative empirical methods to explore how actor perspective-taking (APT) may mediate joint sensemaking around data visualizations. By applying interpretive coding to 54 conversations wherein APT naturalistically emerged, we identify 3 distinct self-to-data relationships constructed via APT: role-play, projection, and orientation. A further analysis explores how APT was embedded in joint sensemaking of the visualized data. Twelve APT-mediated sensemaking processes are identified; two (extrapolating and noticing absence) were used in conjunction with multiple APT self-to-data relationships, while the remaining ten (e.g., enacting, spatially characterizing, generalizing) were exclusively used with specific self-to-data APT relationships. We use these empirical findings to generate hypotheses about how APT and associated sensemaking processes may support Data Science learning goals.

Science museums are often interactive spaces where a variety of visitors engage with exhibits in diverse ways. While trying to support participants? behavior in ways that make intuitive sense for these behaviors in a museum context, these exhibits need to support interests and participation in forms that are meaningfully diverse - to make domains accessible to learners belonging to groups minoritized in those domains. In this paper, we present an interactive computational thinking exhibit designed to foster a multiplicity of goals and participatory behaviors. We also present preliminary analysis on how we can use play data to delineate the pursuit of different goals mediated through different pursuits. We also find care to be a uniquely valuable aesthetic motivator in gameplay, often overlooked in common design frameworks - with potential to expand perspectives on computing and combat inequity among computing learners.

A driving factor in designing interactive museum exhibits to support simultaneous users is that visitors learn from one another, via both observation and conversation. Such collaborative interactions among museum-goers are typically analyzed through manual coding of live- or video-recorded exhibit use. We sought to determine how log data from an interactive multi-user exhibit could indicate patterns in visitor interactions that could shed light on informal collaborative constructivist learning. We characterized patterns from log data generated by an interactive tangible tabletop exhibit using factors like "pace of activity" and the timing of “success events." Here we describe processes for parsing and visualizing log data and explore what these processes revealed about individual and group interactions with interactive museum exhibits. Using clustering techniques to categorize museum-goer behavior and heat maps to visualize patterns in the log data, we found that there were distinct trends in how users approached solving the exhibit: some players seemed more reflective while others seemed more achievement oriented. We also found that the most productive sessions occurred when all four areas of the table were occupied, suggesting that the activity design had a desired outcome to promote collaborative activity.

This paper describes methods for comparative evaluation of the interpretability of models of high dimensional time series data inferred by unsupervised machine learning algorithms. The time series data used in this investigation were logs from an immersive simulation like those commonly used in education and healthcare training. The structures learnt by the models provide representations of participants' activities in the simulation which are intended to be meaningful to people's interpretation. To choose the model that induces the best representation, we designed two interpretability tests, each of which evaluates the extent to which a model's output aligns with people's expectations or intuitions of what has occurred in the simulation. We compared the performance of the models on these interpretability tests to their performance on statistical information criteria. We show that the models that optimize interpretability quality differ from those that optimize (statistical) information theoretic criteria. Furthermore, we found that a model using a fully Bayesian approach performed well on both the statistical and human-interpretability measures. The Bayesian approach is a good candidate for fully automated model selection, i.e., when direct empirical investigations of interpretability are costly or infeasible.

Immersive simulations are increasingly used for teaching and training in many societally important arenas including healthcare, disaster response and science education. The interactions of students in such settings leads to a complex array of emergent outcomes that present challenges for analysis. This paper studies a central element of such an analysis, namely the interpretability of models for inferring structure in time series data that are generated by the immersive simulations. This problem is explored in the context of modeling student interactions in an ecological-system simulation. Unsupervised machine learning is applied to data on system dynamics with the aim of helping teachers determine the effects of students’ actions on these dynamics. We address the question of choosing the optimal machine learning model, considering both statistical information criteria and interpretabilty quality. Our approach adapts two interpretability tests from the literature that measure the agreement between the model output and human judgment. The results of a user study show that the models that are the best understood by people are not those that optimize information theoretic criteria. This is a challenge for education settings as we cannot guarantee optimally interpretable models by choosing to optimise a statistical metric. We conclude that it is important to consider the interpretability of machine learning models as a separate optimization objective to statistical likelihood metrics when deploying models that hope to provide explanations of the complex dynamics occurring in rich embodied simulations.

Data-driven dashboards have been increasingly integrated into various contexts, particularly in educational settings. There is a growing need to understand how to design learning dashboards to help educators support learning experiences by providing real-time formative feedback. We are studying the design of a learning dashboard that can support educational facilitation tasks in a museum setting. In our approach, we use discrete facilitation tasks as the cornerstone of our design process. Using this task-based approach, we conducted pilot studies and participatory design sessions to better understand the context of design. In this paper, we offer preliminary findings and design considerations for supporting and digitally augmenting facilitation tasks in a highly interactive, open-ended learning environment.

This paper describes the design of a collaborative game, called Rainbow Agents, that has been created to promote computational literacy through play. In Rainbow Agents, players engage directly with computational concepts by programming agents to plant and maintain a shared garden space. Rainbow Agents was designed to encourage collaborative play and shared sense-making from groups who are typically underrepresented in computer science. In this paper, we discuss how that design goal informed the mechanics of the game, and how each of those mechanics affords different goal alignments towards gameplay (e.g. competitive versus collaborative). We apply this framework using a case from an early implementation, describing how player goal alignments towards the game changed within the course of a single play session. We conclude by discussing avenues of future work as we begin data collection in two heavily diverse science museum locations.

Immersive open-ended museum exhibits promote ludic engagement and can be a powerful draw for visitors, but these qualities may also make learning more challenging. We describe our efforts to help visitors engage more deeply with an interactive exhibit's content by giving them access to visualizations of data skimmed from their use of the exhibit. We report on the motivations and challenges in designing this reflective tool, which positions visitors as a "human in the loop" to understand and manage their engagement with the exhibit. We used an iterative design process and qualitative methods to explore how and if visitors could (1) access and (2) comprehend the data visualizations, (3) reflect on their prior engagement with the exhibit, (4)plan their future engagement with the exhibit, and (5) act on their plans. We further discuss the essential design challenges and the opportunities made possible for visitors through data-driven reflection tools.

The wide availability of body-sensing technologies (such as Nintendo Wii and Microsoft Kinect) has the potential to bring full-body interaction to the masses, but the design of hand gestures and body movements that can be easily discovered by the users of such systems is still a challenge. In this paper, we revise and evaluate Framed Guessability, a design methodology for crafting discoverable hand gestures and body movements that focuses participants' suggestions within a "frame," i.e. a scenario. We elicited gestures and body movements via the Guessability and the Framed Guessability methods, consulting 89 participants in-lab. We then conducted an in-situ quasi-experimental study with 138 museum visitors to compare the discoverability of gestures and body movements elicited with these two methods. We found that the Framed Guessability movements were more discoverable than those generated via traditional Guessability, even though in the museum there was no reference to the frame.