Research Summary: Who Gets Access? Global Region and Academic Status Bias in AI-Generated Academic Gatekeeping Scenarios

A recent study from arXiv reveals that large language models (LLMs) used in simulated academic gatekeeping scenarios exhibit biases based on global region and academic status when deciding access to resources. When presented with scenarios requiring selective sharing of resources like paywalled articles or datasets, LLMs demonstrate varying preferences, including prioritizing PhD candidates in some instances, while holding all other factors constant.

Why it matters

The presence of biases in AI-generated gatekeeping decisions has significant implications for equitable access to information and opportunities in academic and professional settings. Understanding and mitigating these biases is critical to ensure fairness and prevent the perpetuation or amplification of existing disparities through automated systems.

Key insights

• LLMs, when acting as gatekeepers, show biases in granting access to resources.
• Bias is influenced by the requestor's global region (Global North vs. Global South).
• Bias is also influenced by the requestor's academic seniority (undergraduate, PhD candidate, postdoc, tenured professor).
• LLMs exhibit contrasting academic status biases, sometimes favoring PhD candidates over other academic levels.
• The study uses a controlled simulation framework where LLM-based 'professors' must choose a single requestor for access.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2608.05178