New paper claims latent diffusion models embed market logic
A new academic paper argues that diffusion models like Stable Diffusion operate as 'neural economies' that commodify social communication.
Researcher Eryk Salvaggio’s paper *The Market in the Model: Latent Diffusion as Neural Economy* challenges conventional critiques of generative AI by arguing that diffusion models like Stable Diffusion don’t just reproduce biases in their training data—they actively embed market logics into their core mechanisms.
Salvaggio analyzes the technical components of latent diffusion models (LDMs) and traces how each step in the training and generation pipeline converts social communication into abstracted, market-compatible vectors. He introduces the concept of 'neural exchange value,' arguing that these models function as 'neural economies' that commodify imagery by aligning visual generation with platform and attention economy incentives. The paper warns that traditional critiques focused on copyright or commodity fetishism risk reinforcing the very logics these models perpetuate, instead advocating for analyses centered on social exchange and the displacement of human meaning-making.
- Diffusion models like Stable Diffusion embed 'neural economies' that commodify social communication into market-compatible formats.
- Salvaggio analyzes how training pipelines convert visual generation into abstracted, exchangeable vectors aligned with platform incentives.
- Critiques focused solely on copyright or data sourcing may overlook how technical mechanisms reinforce economic logics.
Why It Matters
This challenges how we critique AI-generated content by exposing how market logic is structurally embedded in the models themselves.