Exploring, Co-Creating, and Reflecting on Embodied Creative Dance Computing Learning Activities for AI Education

• In Event: Integrating AI Education in K–12 Humanities Curricula: Opportunities, Challenges, Design Considerations, and Promising Solutions

Abstract

Objectives or purposes: There has been an influx of attention on how to teach AI while engaging a wide variety of learners, with working groups organizing to put together frameworks to outline what constitutes AI Literacy (Long & Magerko, 2020) and how AI competencies can integrate into K12 education (AI4K12, n.d.). Contributing to this line of research, our work focuses on understanding how to create culturally sustaining experiences for learners to explore AI and machine learning (ML) concepts through embodied dance practices.

Theoretical framework: We implemented co-design as a method for facilitating and centering instructors’ values, ownership, and authentic contexts in the realization, implementation, and evaluation of learning designs for students (Roschelle et al., 2006). We used the AI4K12 guidelines (AI4K12, n.d.) to guide sessions with dance teachers as we engaged in co-designing activities for learners across dance and ML.

Methods and Data Sources: Three dance education teachers participated in two co-design sessions on designing learning activities that support embodied AI education through dance. In the first session, teachers explored ML models with danceON (Payne et al., 2021), a web-based, open-access creative coding environment that uses pose detection algorithms and enables users to draw virtual animations over user-produced videos. Using an ML model pre-trained through GTM and uploaded into danceON, teachers explored and discussed the model’s behavior in relation to different body movements and cases when it failed to recognize poses. Teachers then trained their own dance pose models, experiencing the processes of gathering pose data, training a model on the data, and exporting their model into a live system (i.e., danceON). In the second session, teachers simulated the neural network ML algorithm to get a sense of its inner workings—teachers took the roles of “neurons” to classify various pose images, then discussed their understanding of the algorithm, scenarios in which it incorrectly classified images, and how its limitations could impact users of danceON and similar AI-powered software. Drawing on these discussions and the design session activities, teachers designed learning activities for their students across dance and ML using danceON (and/or GTM).

Results: During co-design sessions, teachers discussed their experiences around AI and its opportunities and limits with their own students. In designing learning activities, teachers thought about the dance and AI concepts they are teaching and how to create space to bring students’ prior knowledge and experiences into the activities. The teachers created activities that deeply intertwined dance practices with ML concepts, for example—incorporating ideas around emotion (excited vs. happy) and movements (axial vs. locomotor) with collecting data samples of its various versions for these to be recognized; and creating movement sentences with danceON and GTM to be transferred into a live performance and thinking about how the technology influenced the dance.

Significance: Our findings demonstrate how co-designing with teachers can lead to learning activities that meaningfully intertwine embodied and creative dance practices with ML concepts. The co-design sessions provided a way to support teachers in thinking about how their dance discipline and teaching experiences can be used to teach AI within their practice.

Authors

• Francisco Castro, New York University

• Kayla DesPortes, New York University