https://taxonomy.eticas.ai/risk/homogenization-output-across-groups
Maturity: emerging
The system produces outputs that flatten cultural, linguistic, or stylistic diversity, defaulting to dominant patterns and erasing distinctive characteristics of underrepresented groups. Particularly relevant for generative AI, where homogenization in text, image, or speech outputs can reinforce hegemonic norms.
This subcategory is emerging. It has not yet been validated through established assessment methods.
System type: Large language models (LLM)
Lifecycle stages: Post Processing
| Framework | Reference |
|---|---|
| NIST AI 600-1 — Generative AI Risk Profile | Harmful Bias or Homogenization (homogenization clause) |
| Framework | Reference |
|---|---|
| MIT AI Risk Repository | Economic and cultural devaluation of human effort |
| IBM AI Risk Atlas | Output diversity / mode collapse |