Anti-AI Bias Toward Couple Images and Couple Counseling: Findings from Two Experiments

Area of Science:

• Psychology
• Human-Computer Interaction
• Artificial Intelligence Ethics

Background:

• Generative artificial intelligence (AI) systems are increasingly used in sensitive domains like intimacy and sexuality.
• Skepticism towards AI-generated content is prevalent, leading to potential anti-AI bias.
• Anti-AI bias involves systematically devaluing AI outputs compared to human-created content.

Purpose of the Study:

• To investigate if AI-generated couple images and counseling dialogues are perceived less favorably than human-created equivalents.
• To examine if individual AI attitudes and AI literacy moderate these perceptions.
• To extend the understanding of anti-AI bias into intimate and sexual contexts.

Main Methods:

• Two online experiments (N=2,658) with a German adult sample in 2024.
• Utilized an experimental labeling paradigm, comparing evaluations of identical content labeled as AI-generated versus human-created.
• Measured AI attitudes and AI literacy as potential moderators.

Main Results:

• AI-generated couple images were rated less positively than human-generated images (d=.21).
• AI-generated counseling dialogues were evaluated less favorably than human-generated dialogues (d=.23).
• AI attitudes and literacy moderated the bias effect for images, but not for counseling dialogues.

Conclusions:

• Findings confirm the existence of anti-AI bias in intimate contexts for both visual and textual content.
• User dispositions towards AI play a role in moderating bias, particularly for visual content.
• Highlights the need to consider user psychology when deploying AI in sensitive areas like sexual health and relationships.