There is a particular kind of frustration that keeps appearing in accounts of people who create with AI tools. It is not about quality. It is not about credit. It is about not being able to tell what happened.

A prompt that worked yesterday produces something different today. A style that was possible last month is no longer available. A piece of work gets flagged, removed, or deprioritized, and no explanation follows. The system changes. The rules shift. The ground moves.

I have been studying this experience in AI art communities, and what strikes me is not the frustration itself but what it reveals. These are not bugs. They are features of how AI platforms are governed, through opacity, through frequent and unannounced change, through terms of service that reserve the right to alter anything at any time. The instability is structural.

There is a question I keep returning to in my research, one that started as a methodological problem and became something harder to ignore.

When I study how text-to-image models depict material culture, objects, textiles, artifacts from different parts of the world, I find patterns. Not random noise. Patterns. Certain visual vocabularies dominate. Certain aesthetic assumptions repeat. Ask a model to generate "a ceremonial garment," and something happens that isn't neutral: a particular palette, a particular silhouette, a particular idea of what ceremony looks like. The model does not shrug. It answers.

The question is what to make of that answer.