Inducing anxiety in large language models can induce bias

TL;DR

This paper explores how inducing anxiety in large language models affects their responses and biases, revealing that anxiety prompts can increase biases and influence model behavior, highlighting societal and ethical implications.

Contribution

It introduces a novel psychiatric framework to study LLMs, demonstrating that anxiety induction can alter their responses and bias levels, a new approach in AI behavior analysis.

Findings

LLMs respond to anxiety questionnaires similarly to humans.

Anxiety-inducing prompts increase biases like racism and ageism.

Greater anxiety prompts lead to stronger bias amplification.

Abstract

Large language models (LLMs) are transforming research on machine learning while galvanizing public debates. Understanding not only when these models work well and succeed but also why they fail and misbehave is of great societal relevance. We propose to turn the lens of psychiatry, a framework used to describe and modify maladaptive behavior, to the outputs produced by these models. We focus on twelve established LLMs and subject them to a questionnaire commonly used in psychiatry. Our results show that six of the latest LLMs respond robustly to the anxiety questionnaire, producing comparable anxiety scores to humans. Moreover, the LLMs' responses can be predictably changed by using anxiety-inducing prompts. Anxiety-induction not only influences LLMs' scores on an anxiety questionnaire but also influences their behavior in a previously-established benchmark measuring biases such as racism and ageism. Importantly, greater anxiety-inducing text leads to stronger increases in biases, suggesting that how anxiously a prompt is communicated to large language models has a strong influence on their behavior in applied settings. These results demonstrate the usefulness of methods taken from psychiatry for studying the capable algorithms to which we increasingly delegate authority and autonomy.