AI Application Gives Users Real-Time Feedback on the Level of Peace in the Social Media Videos They Watch

Researchers have built a Chrome extension that uses large language models to give YouTube viewers real-time feedback on how peaceful the videos they watch are, according to a paper posted to arXiv [1]. The application, called MirrorMirror, analyzes the transcript of a YouTube video and measures five social dimensions tied to peace identified across 59 social science studies: compassion-contempt, news-opinion, promotion-prevention, creativity-order, and nuance-simplification [1][2]. The tool is designed to make users more aware of the emotional tone of their media diet and to help content creators understand the character of their own work [3]. The research team, which includes Larry Liebovitch, first attempted to train neural networks on written news text tagged with country-level peace surveys. One fully connected feedforward network and two convolutional neural networks reached approximately 97% accuracy on a test set and roughly 70% on a separate news dataset, but the models did not generalize to YouTube transcripts, indicating that written text differs from transcribed spoken language [1][2]. For YouTube, where no external peace-tagging data exists, the team compared two word-level sentiment analysis tools against six context-level large language models. Three human coders evaluated 52 videos, and the LLMs achieved a correlation coefficient of r² ~ 0.60 with the human judgments, while sentiment analysis reached only r² ~ 0.03 [1][2]. The extension was developed through Human Centered Design, with iterative feedback from team members, design students, and a broader group of testers [4]. The long-term goal is for MirrorMirror to become an open-source tool for content creators, journalists, researchers, and platforms [4][5]. The work arrives as social media’s role in shaping public discourse faces continued scrutiny. A 2024 United Nations committee report noted that Western social media platforms disproportionately removed content expressing solidarity with Palestinians relative to content promoting violence [8]. The New York Times has described the start of the Gaza war as releasing a “deluge of online propaganda and disinformation” larger than anything seen before [8]. Content creators on platforms such as YouTube and Facebook often design videos to increase emotional activation, particularly anger, to drive clicks, the paper notes [5]. The researchers write that “such self-awareness may be a helpful tool in turning down the destructive, polarizing, heat in social media communications” [4]. The extension is available for download at bait.fyi [3].