Pastiche, place and peripheries: algorithmic visions of Australian small towns and cities

This study is interested in how generative AI (GenAI) systems visualise a particular kind of place: small-town communities in Australia and landmarks in or around these communities. When GenAI systems – themselves often made in foreign countries and trained on heterogenous data sources – are asked to visualise these places, what can we learn about how these systems work and the implications of this for both local residents and geographically distant audiences, including those in urban Australia and abroad? To explore this question, we ask five popular GenAI systems (ChatGPT, CoPilot, Firefly, Gemini, and Meta AI) to visualise six small towns in Australia and four local landmarks in or around each of these small towns. We then undertake a semiotic analysis to systematically evaluate the images’ modality (from impressionistic styles through to photorealistic styles), aspect ratios, vertical perspective, and level of multi-modality (such as whether the output includes any written text or symbols) to explore how the AI systems try to convey far-flung places with smaller amounts of available training data. Doing so allows us to explore how much of the storytelling is done by showing local recognisable landmarks versus using signs or words to tell or ‘label’ the location ostensibly being described. We also, using the theoretical lens of the pastiche, reflect on how ‘accurate’ the AI systems are in their renderings and muse about the implications of this for local community members, non-residents, and scholars of visual communication and media.