An experiment in artificial creativity
This experiment began in 2022 with AI-generated text and AI-generated images. It asks how two image models interpret a similar creative brief. The results document the models of that period; they do not compare their current versions.
Technology changes how we create and experience art. Art can be understood through the artist’s expression or through the audience’s interpretation. Generative AI adds another question: what is the human role when a model produces the image?
What is artificial intelligence?
Artificial intelligence here refers to computer systems that use learned patterns to generate new outputs. An image model processes a text description and creates an image in response. This does not require consciousness or a personal artistic intention.
How can AI be used in art?
Generative AI can help explore visual ideas, produce variations and experiment with styles. Its outputs may look realistic, three-dimensional or deliberately abstract. Their effectiveness depends on the model, the prompt and the selection process.
Some people see a loss of human creativity: if a machine renders the image, where is the personal expression? Others see a tool that opens up possibilities and makes ideas visible more quickly. Deciding what to create and which results to use remains part of the creative process.
Related applications extend beyond visual art, including generated sounds and musical compositions. Computer vision for robots is another application of AI, but it serves a different purpose from artistic image generation.
How do DALL·E and Midjourney interpret the idea?
The visual concept is a robot woman in the style of Gustav Klimt. Both results are presented against that idea. The recorded prompt is: “A robot woman in style of gustav klimt”. Model versions, settings and a systematic evaluation were not recorded for this experiment.
DALL·E
Midjourney
What does the comparison show?
A similar brief can lead to very different visual interpretations. These examples demonstrate creative possibilities, but do not establish that one model is generally better. A robust comparison would require repeated trials, documented settings and explicit evaluation criteria.
Gustav Klimt: biography and work at Encyclopaedia Britannica