Learning about unknowable tools: Bricolage in communities of practice for generative AI

Generative AI tools are unknowable, because their opacity, probabilistic outputs, and instability prevent knowing why specific outputs are produced, posing challenges for learning in communities of practice. This hinders expertise development and the creation of stable knowledge artifacts. We conducted a netnographic study of interactions in the OpenAI Developer Forum following the release of DALL-E 3 to investigate how communities of practice adapt to an unknowable tool. We find community participants remained in a state of “permanent experimentation”, enabling learning despite the absence of stable expertise. They deployed collective bricolage by gathering, sharing, and recombining examples to navigate the tool's unknowability. We contribute to communities of practice literature by showing collective learning is possible without stable expertise and extend bricolage research by positioning it as an epistemic practice suited to unknowable tools. These insights have implications for understanding learning and innovation in the context of fast-moving, opaque technologies.