Concerns on Bias in Large Language Models when Creating Synthetic Personae

TL;DR

The paper investigates bias and customization challenges when using synthetic personae in HCI with LLMs, emphasizing the need for thorough testing due to the black-box nature and potential for manipulation. It presents a sub-study using contextual vignettes and adversarial prompts to elicit and examine bias in LLM interfaces. Key contributions include a pilot vignette methodology, an analysis of model-specific susceptibilities to prompting, and a call for ethical, feminist- and queer-informed HCI practices. The work lays a foundation for responsible design and outlines future directions such as non-human participant workshops and speculative provotyping to move beyond current LLM limitations.

Abstract

This position paper explores the benefits, drawbacks, and ethical considerations of incorporating synthetic personae in HCI research, particularly focusing on the customization challenges beyond the limitations of current Large Language Models (LLMs). These perspectives are derived from the initial results of a sub-study employing vignettes to showcase the existence of bias within black-box LLMs and explore methods for manipulating them. The study aims to establish a foundation for understanding the challenges associated with these models, emphasizing the necessity of thorough testing before utilizing them to create synthetic personae for HCI research.