The imagined affordances of algorithms astechnologies of representation

Research question My research addresses a broad question -how are algorithms imagined to help us see and know the world? -by examining their growing use as tools for producing texts. Algorithms have begun to capture the public imagination as a way to produce texts as diverse as paintings, screenplays, and music. Algorithms produce texts (understood as any form of media) by applying predefined rules to a data set. At stake is a way of representing the world according to a procedural logic, which we might counterpose against an editorial logic based on the subjective interpretations of an expert. My proposed project examines the appeal of producing texts in this way, asking why we might turn to an algorithm to represent the world rather than a human. My research focuses on three case studies. They interrogate three classes of algorithm -fixed, learning, and 'black boxed' -in relation to three diverse objects of concern: the environment, culture, and the self. The first case is Ed Hawkins' #warmingstripes campaign, where an open-source algorithm converts temperature data into coloured stripes ranging from blue (cold) to red (hot), thereby presenting patterns of climate change that are otherwise invisible. The second case focuses on the 'I forced a bot' meme, which presents the results of a neural network that has been trained to reimagine texts from popular culture, appearing to show how an algorithm sees human culture. The third case is the Swiftkey game on Reddit, where an autocomplete algorithm draws on personal data to anticipate what users will say next and therefore tell them something about themselves. The texts in each case claim to reveal a truth through algorithmic representation of data. Rather than focusing on any single algorithm or dataset, my research combines these case studies to interrogate a more fundamental question about the perceived benefits of algorithmic representation. I approach these benefits as 'imagined affordances': features that are...