We’re excited to share two connected papers investigating bias in AI-recommended scholars, both featuring group members Fariba Karimi and Lisette Espín-Noboa!
In the first, Daniele Barolo, Chiara Valentin, Fariba Karimi, Luis Galárraga, Gonzalo G. Méndez, and Lisette Espín-Noboa audited six open-weight LLMs on their ability to recommend physics experts, finding systematic biases toward male, highly-cited, and geographically concentrated scholars.
The follow-up, by Lisette Espín-Noboa and Gonzalo G. Méndez, expanded the audit to 22 LLMs and tested common fixes — raising the model’s “temperature,” constraining prompts for representation, and grounding answers with web search (RAG). None of these turned out to be a free lunch: each intervention improves some outcomes while making others worse, showing that fixing bias in AI recommendations isn’t as simple as flipping a setting.
The work picked up real attention: a Complexity Science Hub feature, press coverage in Infobae and APA Science, and an interactive visualization letting you explore which scholars different LLMs surface.
📖 Read Paper I — Barolo et al., 2025
📖 Read Paper II — Espín-Noboa & Méndez, KDD ’26
📰 CSH news feature
🌐 Try the interactive visualization: “Whose name comes up?”
📰 Coverage in Infobae
Image by همَّام on Unsplash
Network Inequality and Fairness Group 
Leave a comment