The World We Made

Prompt-based AI Generated Imagery • 2023

A series of images made with MidJourney that reflect on the nature of AI media and image making

The recent generation of text-to-image systems challenges us to reconsider what an "image" really is, and to ask if AI is significantly changing the meaning of "image making" and the creative practices built around that (e.g. photography, painting, illustration, cinema...even "traditional" computer generated imagery).

The rise of prompt-based generative AI raises many cultural, conceptual and ethical dilemmas, not least that these systems are trained on a large corpus of human art, without express permission from the artist creators themselves.

This series of images were made using short prompts without detailed reference to the kind of people represented in the images, for example words like "woman", "grandmother", "teenager", "mother with baby" and “middle aged man” were used. The system then generates images that represent the statistical regression of that term’s visual concept according to the training data: what the collective internet considers the typical image of each word-object association.

It's easy to give the images produced a faux sense of high production values, professionalism, editorial style, even a kind of sincerity. However, on closer inspection, errors in the system begin to reveal themselves, particularly when dealing with reflections and space – providing some insight into how the system constructs a coherent image. This reminds us that these are not photographs, and they do not depict living persons.

In the sequence below prompts such as "mother with baby" and "woman brestfeeding" were used:

Closer inspection reveals the current inability of generative AI to create complete coherence. Skin displays a telltale waxy appearance, faces are distorted, hands and other body-parts are malformed.

‌The figures seem to be lost in some kind of endless gallery of mirrors, trying to photograph themselves with their smartphones (perhaps to see if they are really recorded by the camera's gaze?) Some are depicted absorbing themselves into another virtual world, to hide from the uncomfortable sterility of their AI generated setting. Many "see" themselves in a reflection that's not quite right. The framing suggests a visual ambiguity – are the figures depicted looking at some imaginary "photographer" taking their picture, the viewer (you), or is the viewer positioned as one more mirror and the figures, looking into that mirror, are wondering about their own reflection, perhaps taking a selfie to immortalise the moment forever or to validate their own existence?

Generative AI's technical ability to automate image production must surely represent a step-change in how and why we make images. Each image is generated in just a few seconds and the supply is practically endless, but the images produced are always generic in nature, parasitic on human culture and art, never innovative or transformational.

This is the world we made, where machines provide us with a visceral and divisive mirror on our own cultural and technical excess.

Further reading:
Jon McCormack, Camilo Cruz Gambardella, Nina Rajcic, Stephen James Krol, Maria Teresa Llano and Meng Yang (2023). Is Writing Prompts Really Making Art? (C. Johnson, N. Rodríguez-Fernández, & S. M. Rebelo, Eds.), Artificial Intelligence in Music, Sound, Art and Design, pp. 196-211. Cham: Springer Nature Switzerland.
Arxiv preprint.