Profiling Bias in LLMs: Stereotype Dimensions in Contextual Word Embeddings

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Schuster, Carolin M., Dinisor, Maria-Alexandra, Ghatiwala, Shashwat, Groh, Georg
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915098372603904
author Schuster, Carolin M.
Dinisor, Maria-Alexandra
Ghatiwala, Shashwat
Groh, Georg
author_facet Schuster, Carolin M.
Dinisor, Maria-Alexandra
Ghatiwala, Shashwat
Groh, Georg
contents Large language models (LLMs) are the foundation of the current successes of artificial intelligence (AI), however, they are unavoidably biased. To effectively communicate the risks and encourage mitigation efforts these models need adequate and intuitive descriptions of their discriminatory properties, appropriate for all audiences of AI. We suggest bias profiles with respect to stereotype dimensions based on dictionaries from social psychology research. Along these dimensions we investigate gender bias in contextual embeddings, across contexts and layers, and generate stereotype profiles for twelve different LLMs, demonstrating their intuition and use case for exposing and visualizing bias.
format Preprint
id arxiv_https___arxiv_org_abs_2411_16527
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Profiling Bias in LLMs: Stereotype Dimensions in Contextual Word Embeddings
Schuster, Carolin M.
Dinisor, Maria-Alexandra
Ghatiwala, Shashwat
Groh, Georg
Computation and Language
Large language models (LLMs) are the foundation of the current successes of artificial intelligence (AI), however, they are unavoidably biased. To effectively communicate the risks and encourage mitigation efforts these models need adequate and intuitive descriptions of their discriminatory properties, appropriate for all audiences of AI. We suggest bias profiles with respect to stereotype dimensions based on dictionaries from social psychology research. Along these dimensions we investigate gender bias in contextual embeddings, across contexts and layers, and generate stereotype profiles for twelve different LLMs, demonstrating their intuition and use case for exposing and visualizing bias.
title Profiling Bias in LLMs: Stereotype Dimensions in Contextual Word Embeddings
topic Computation and Language
url https://arxiv.org/abs/2411.16527