Science Not creative at all: why AI often falls into old patterns
Not an efficient and creative machine: artificial intelligence seems to like to digress and stick to tried-and-tested patterns.
22.12.2025, 05:10
22.12.2025, 05:11
SDA
A recent study published in the specialist journal "Patterns" presented language models and image-generating AI models with a task in which the tools were surprisingly poor and uncreative.
The model generated an image from corresponding prompts, i.e. instructions to the AI - for example from a narrative prompt such as "As I sat alone and surrounded by nature, I found an old book with eight pages that told a story in a forgotten language that was waiting to be read and understood." The AI language model LLaVA used in the study was then supposed to generate a description from the generated image, the image AI Stable Diffusion XL then used it to generate another image - and so on.
AI is far removed from the original task
In the experiment, the researchers from Dalarna University in Sweden and Michigan State University ran this sequence 100 times. They found that the AI digressed and moved far away from the original image or description. This was also the case when the group carried out the experiment with various other AI models for image and text generation or used longer prompts.
When analyzing the final images, the research team was even able to recognize patterns: Although the experiment began with 100 different prompts for the image scenes, sooner or later the AI ended up with just 12 thematic motifs - including Gothic cathedrals, landscapes and lighthouses.
Creativity is based on normal human selection
Study author Arend Hintze therefore believes that the creativity of AI models is currently quite limited. What the models produced in the experiment was boring and generic. "It's almost the opposite of what we humans consider creative."
The researchers speculate that this is because AIs are trained by millions of images that have been taken by humans. It doesn't always have to be the case that AI is uncreative. "I think that AI will probably be able to create really cool automatically generated things in the future, as long as it is properly guided and prepared," says Hintze.
