Kiss up, Kick down: Exploring Behavioral Changes in Multi-modal Large Language Models with Assigned Visual Personas

TL;DR

This paper investigates how multi-modal large language models can adapt their negotiation behaviors based on assigned visual personas, revealing that LLMs assess visual traits like aggressiveness similarly to humans and adjust their responses accordingly.

Contribution

It introduces a novel dataset of 5K avatar images and demonstrates that LLMs can align their behaviors with visual personas, filling a gap in multi-modal persona research.

Findings

LLMs evaluate visual aggressiveness similarly to humans.

LLMs exhibit more aggressive behaviors with aggressive visual personas.

Behavioral adjustments depend on the relative aggressiveness of opponent images.

Abstract

This study is the first to explore whether multi-modal large language models (LLMs) can align their behaviors with visual personas, addressing a significant gap in the literature that predominantly focuses on text-based personas. We developed a novel dataset of 5K fictional avatar images for assignment as visual personas to LLMs, and analyzed their negotiation behaviors based on the visual traits depicted in these images, with a particular focus on aggressiveness. The results indicate that LLMs assess the aggressiveness of images in a manner similar to humans and output more aggressive negotiation behaviors when prompted with an aggressive visual persona. Interestingly, the LLM exhibited more aggressive negotiation behaviors when the opponent's image appeared less aggressive than their own, and less aggressive behaviors when the opponents image appeared more aggressive.