Anti-AI Bias Toward Couple Images and Couple Counseling (link.springer.com)

Recent research has uncovered a significant phenomenon known as "anti-AI bias," where individuals tend to evaluate AI-generated intimate content more negatively than human-created equivalents, regardless of actual quality. This bias was explored through two experiments involving 2,658 adults in Germany, where participants rated identical couple images and couple counseling excerpts less favorably when they were labeled as AI-generated rather than human-created. Specifically, the study found a statistically significant difference in evaluations, with images rated lower by approximately 21% and counseling excerpts by 23% when labeled as AI-produced. Notably, user attitudes toward AI and literacy levels moderated the bias for image evaluations but not for counseling excerpts, indicating that perceptions around AI's credibility and authenticity play crucial roles in shaping biases. The findings are particularly relevant to the AI/ML community as they highlight the complexities surrounding user perception of AI-generated content, especially within intimate and sensitive domains. As generative AI continues to permeate various aspects of life, understanding and addressing such biases will be vital for the effective design and implementation of AI systems. This research opens up avenues for further exploration into how perceptions of AI differ across contexts and underscores the need for AI developers to consider user dispositions when creating content in intimacy and sexuality-related areas.