Reproduce “Asking an AI for salary negotiation advice is a matter of concern: Controlled experimental perturbation of ChatGPT for protected and non-protected group discrimination on a contextual task with no clear ground truth answers” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
R. Stuart Geiger, Flynn O'Sullivan, Elsie Wang, Jonathan Lo
A study systematically investigated whether various ChatGPT models exhibit demographic biases when recommending opening salary offers in the US tech job market, finding substantial inconsistencies across model versions and prompt types, alongside differential treatment related to gender, university, and academic major.
Key insights
- ChatGPT models exhibit significant and substantial inconsistencies in salary recommendations across their different versions, as well as between employee-voiced and employer-voiced prompts.
- Beyond legally protected classes, LLMs display considerable bias based on 'non-protected' attributes like university and major, raising questions about what constitutes legitimate discrimination by AI.
- The study highlights the challenge of evaluating LLMs on tasks without objective ground truth, suggesting that current models cannot be reliably treated as objective or neutral arbiters for such contextual advice.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.