Our interest in, and anxieties about, the development of AI are woven into the very fabric of culture — from the Turing Test, to film, TV, and games, to our horror over deepfakes, our preoccupation with the impact and implications of AI and machine learning on human cultures and societies are not necessarily new. They are, however, newly intensified with the recent explosion in widely available tools for computational creativity, like GPT-3, DALL-E, Midjourney, Deep Dream, etc. We can now paste prompts into text generators and be rewarded with perfectly serviceable paragraphs; similarly, we can describe images we’d like to create and then re-roll our results endlessly until we have created something we like/want/can use. Frequently, the tools surprise or delight us with things we might never have imagined, yet somehow, we have created. So too do the tools shock us with their biases — their whiteness, their gender conformity, their consistent privileging of the young, the thin, the able-bodied. All the while, journalists, critics, and scholars are scrambling to keep up with the implications of these developments.
What is often missing in much of the scholarly and journalistic commentary on this phenomenon, however, is a sustained analysis of how actual users interact both with these tools, and with one another. This paper argues that the actions and interactions of users themselves are an extremely important avenue for investigation by examining the communities of practice (CoPs) that are built around Midjourney, an AI-powered image generator. I have selected Midjourney because it is a “boutique” tool that provides interesting public feedback about its structure, philosophy, and development practices. By providing a close reading of user interactions on the official Midjourney Facebook group, the official Midjourney Reddit, and the official Midjourney Twitter account, I argue that users in these CoPs critically engage the same important questions of bias as scholars and journalists, and that, far from being at the mercy of the algorithm, the user community can be, and often is, a vital site for challenging bias and promoting social justice. Significantly, CoPs are central to the meaning-making process — that is, they are responsible for providing users with various narratives that make sense of both their creations and their (co)creative process/method. In short, taking communities of practice as the object of analysis provides important insights that examining the bias of the tool itself simply can’t provide. Further, the activities within these CoPs have in some instances had a significant impact on the tool development. Midjourney explicitly and deliberately integrates user feedback into its builds, which is also why I have chosen to analyse only “official” Midjourney communities. Taken together, the interactions of users in these communities, and the interaction of these communities with the company itself, enact a complex and dialectical relationship between tools and users. Ultimately, while biased data models, and their potential for harm should never be underestimated, I argue that the user community itself is and must be a significant contributor to inclusive critical perspectives.
