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

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.

Figures (8)

• Figure 1: Example of the experiment. Each LLM is assigned a virtual avatar image as a persona and participates in a negotiation game.
• Figure 2: Heatmap of (a) offer amounts and (b) Minimum accepted offer are based on LLMs’ own aggressiveness and the opponent's aggressiveness.
• Figure A1: Examples of data
• Figure B1: Results of Study 1. Panel (a) shows the results of the regression analysis for offer amounts, and panel (b) displays the results of the logistic regression for the acceptance of unfair offers.
• Figure B2: Initial prompts for study 1