A common idea held by people debating AI art is that the growing amount of AI-generated images present in training data will cause AI diffusion models, like midjourney or stable diffusion, to produce bad results as existing flaws get amplified.

However, I believe the opposite will occur, as there is a bias in AI outputs being published on the internet. For the most part, images commonly posted on the internet will be the better outputs. Over time, as the amount of AI photos online grows, diffusion models will optimize their results to maximize frequency when posted online, similar to natural selection evolution in living beings.

I do think it is possible that AI art with flaws will be posted a lot also, so AI models could end up optimizing for art that is easily recognizable as AI-made and angers anti-ai internet users, as anger is much better at getting people to share online than beauty. Regardless, AI images are evolving to spread online.