Summary of Ask Llms Directly, “what Shapes Your Bias?”: Measuring Social Bias in Large Language Models, by Jisu Shin et al.

by Jisu Shin, Hoyun Song, Huije Lee, Soyeong Jeong, Jong C. Park

First submitted to arxiv on: 6 Jun 2024

GrooveSquid.com Paper Summaries

GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!

Summary difficulty	Written by	Summary
High	Paper authors	High Difficulty Summary Read the original abstract here
Medium	GrooveSquid.com (original content)	Medium Difficulty Summary Large language models (LLMs) can perpetuate social biases shaped by accumulated perceptions across demographic identities. To comprehensively understand these biases, researchers must consider diverse perspectives among identities. Prior studies have evaluated biases by analyzing sentiments towards demographic identities or measuring alignment with stereotypes. These methods are limited in directly quantifying biases at the level of distinct perspectives. This paper proposes a novel strategy to quantify social perceptions and suggests metrics to evaluate social biases within LLMs by aggregating diverse perspectives. Experimental results demonstrate the quantitative measurement of social attitudes in LLMs, examining social perception. The proposed metrics capture the multi-dimensional aspects of social bias, enabling a fine-grained investigation of bias in LLMs.
Low	GrooveSquid.com (original content)	Low Difficulty Summary This paper is about how big language models can be biased against certain groups of people because of the way they were trained on lots of text data. Researchers want to understand why this happens and how to make these models fairer. Right now, there are limitations in how we measure biases, so the authors propose a new way to do it. They test their idea and show that it works by looking at the attitudes expressed in the language models towards different groups of people.