Generating AI Images: Prompting and Human- Machine Imagination

In this presentation we examine visual generative media, mapping how, as visual sociologists, we can begin to better understand the different sociotechnical practices such tech can afford us as social scientists.

AI has penetrated our lives and visual production long before this breakthrough moment where we now ‘see it’. AI has long been a part of many mobile phones with camera features and apps like Shazam. From the micro analysis of AI in our pockets to macro-discussions around the use of AI in political misinformation and the economic phenomenon of “digital sweatshops” across the global South the necessity to understand the social implications AI in sociological terms is yet to be fully defined.

While there is a growing body of literature on algorithmization of society (Airoldi, 2024), critical dataset studies (Thylstrup, 2022) and critical AI studies (Lindner, 2024) we argue that with exceptions of a few recent studies (Laba, 2024, Smits, 2024) the visual sociological discussion of AI and visual sphere is still lagging behind.

As a set of technologies AI is positioned at a critical frontier of human-machine interaction, where multiple ethical issues, issue of aesthetics and creative work arise. As we go we raise the following questions: How much time will it take the machine to learn the culturally specific images given the diversity of humankind? While it is becoming evident that AI will augment and inspire creativity – what kind of human-machine creativity it may be?

Based on autoethnographies and interviews with artists and graphic designers who use different AI visual generative media (ChatGPT and Midjourney) we reflect on how text-image dynamics of prompting unfolds during the image manufacturing or co-imagination. We use various instances of AI “hallucinations” to discuss the way machines learn “to see” to predict and to surprise us with the visual outputs.