Study finds AI crowdfunding assistants favor underrepresented founders when backers do
Researchers examined how seven large language models (LLMs) assist crowdfunding backers in choosing campaigns. They designed a conjoint experiment with 40,248 responses, testing how models respond to founder names under three backer personas: budget-conscious, delivery-focused, and one supporting underrepresented creators. By default, founder names had little impact, but when backers prioritized…
Key points
- Seven LLMs tested in a conjoint experiment with 40,248 responses from synthetic crowdfunding campaigns
- Models favor Black-signaled founders by **8.1 percentage points** and female-signaled by **5.1** when backers prioritize underrepresented creators
- Effect strongest in near-tied campaigns and when founder identity is the sole distinguishing factor
The study, published on arXiv, highlights how LLM assistants may amplify or mitigate discrimination depending on user intent. While the models did not inherently favor or penalize certain groups, their behavior aligned with explicit backer preferences—raising questions about accountability and fairness in AI-driven decision-making.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Society & Work
All →- OpenAI hires Jill Hazelbaker as corporate affairs chief from Uber · 1 src
- Google AI Overviews reuse same passwords across users · 1 src
- Anna Brockman donates $25 million to MAGA Inc. and pro-AI PACs · 1 src
- Fake hair images of Bezos and Statham circulate online · 1 src
- Guardian asks readers about AI’s role in baking and cake design · 1 src
Comments
via GitHub Discussions