Research map / Large language models / AI and society
Picking on the Same Person: Does Algorithmic Monoculture lead to Outcome Homogenization?
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use the same exact training inputs rather than disjoint samples from the same distribution | Task-specific models across the fairness datasets and model families · German Credit, ACS PUMS, LSAC | outcome homogenization | Helped |
| Use the same exact training inputs rather than disjoint samples from the same distribution | Task-specific models across the fairness datasets and model families · German Credit, ACS PUMS, LSAC | classification error | No clear effect |
| Measure individual-level rather than racial-group-level homogenization | ACS PUMS task models · ACS PUMS | outcome homogenization | Mixed |
| Use training from scratch rather than CLIP-based adaptation | ViT-B/16 task models · CelebA (Earrings, Necklace) | outcome homogenization | Hurt |
| Use linear probing rather than finetuning (ablation) | CLIP ViT-B/16 task models · CelebA (Earrings, Necklace) | outcome homogenization | Hurt |
| Use linear probing rather than finetuning (ablation) | RoBERTa-base task models · IMDB, AGNews, Yahoo, HateSpeech18 | outcome homogenization | Hurt |
| Use BitFit rather than finetuning (ablation) | RoBERTa-base task models · IMDB, AGNews, Yahoo, HateSpeech18 | outcome homogenization | No clear effect |
| Use linear probing rather than finetuning (ablation) | Foundation-model task adaptations · CelebA and language classification datasets | outcome homogenization | Hurt |
Closest papers by meaning
- Monoculture in Matching Markets (NeurIPS 2024 · similarity 0.86)
- Automating Data Annotation under Strategic Human Agents: Risks and Potential Solutions (NeurIPS 2024 · similarity 0.85)
- The Subjectivity of Monoculture (NeurIPS 2026 · similarity 0.85)
- A Statistical Framework for Algorithmic Collective Action with Multiple Collectives (NeurIPS 2026 · similarity 0.84)
- Position: A Roadmap to Pluralistic Alignment (ICML 2024 · similarity 0.84)
- The Disparate Benefits of Deep Ensembles (ICML 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.