Teaching Machine Learning as Adaptive Design Work

To broaden and improve machine learning (ML) education, researchers must partner with teachers and study the affordances and challenges of ML instruction.Yet, most teachers are not ML experts and may need curricula that align ML concepts with their disciplinary expertise.This study explores three cases of classroom enactments of Mathematics of Facial Recognition (FaceID): five co-designed ML lessons aligned to high school statistics standards.Each teacher was encouraged to adapt FaceID to meet their instructional goals.Drawing from Brown's Design Capacity for Enactment framework, we examine the frequency and substance of teachers' adaptations of FaceID, and the design resources that enabled their adaptations.We found that teachers' prior knowledge of ML, their statistics pedagogical content knowledge, and their knowledge of students supported their adaptations.From these findings, we illustrate triumphs and challenges that teachers who are new to ML may face when enacting ML lessons.