A Game-Based Approach for Helping Designers Learn Machine Learning Concepts
Machine Learning (ML) is becoming more prevalent in the systems we use daily. Yet designers of these systems are under-equipped to design with these technologies. Recently, interactive visualizations have been used to present ML concepts to non-experts. However, little research exists evaluating how designers build an understanding of ML in these environments or how to instead design interfaces that guide their learning. In a user study (n=21), we observe how designers interact with our interactive visualizer, \textit{QUBE}, focusing on visualizing Q-Learning through a game metaphor. We analyze how designers approach interactive visualizations and game metaphors to form an understanding of ML concepts and the challenges they face along the way. We found the interactive visualization significantly improved participants' high-level understanding of ML concepts. However, it did not support their ability to design with these concepts. We present themes on the challenges our participants faced when learning an ML concept and their self-guided learning behaviors. Our findings suggest design recommendations for supporting an understanding of ML concepts through guided learning interfaces and game metaphors.