A Matter of Perspective(s): Contrasting Human and LLM Argumentation in Subjective Decision-Making on Subtle Sexism

TL;DR

This paper compares how humans and large language models consider different social perspectives in subjective judgments about subtle sexism, revealing both similarities and differences in their argumentation patterns.

Contribution

It introduces a classification of perspectives in subjective decision-making and analyzes how LLMs differ from humans in perspective distribution and combination.

Findings

Perspectives are classified into perpetrator, victim, decision-maker.

Humans and LLMs show different distributions of perspectives.

Systematic evaluation of LLMs' perspective-taking is necessary.

Abstract

In subjective decision-making, where decisions are based on contextual interpretation, Large Language Models (LLMs) can be integrated to present users with additional rationales to consider. The diversity of these rationales is mediated by the ability to consider the perspectives of different social actors. However, it remains unclear whether and how models differ in the distribution of perspectives they provide. We compare the perspectives taken by humans and different LLMs when assessing subtle sexism scenarios. We show that these perspectives can be classified within a finite set (perpetrator, victim, decision-maker), consistently present in argumentations produced by humans and LLMs, but in different distributions and combinations, demonstrating differences and similarities with human responses, and between models. We argue for the need to systematically evaluate LLMs' perspective-taking to identify the most suitable models for a given decision-making task. We discuss the implications for model evaluation.