ACM HyperText 2025 conference accepted paper: “Synthetic Politics: Prevalence, Spreaders, and Emotional Reception of AI-Generated Political Images on X”

Artificial Intelligence isn’t just shaping the future of technology; people’s perceptions are on the line with its vast impact. In many instances, we don’t notice how political images affect our emotions, and it is very simple for AI images to exploit our trust in democracy, blurring the line of truth and manipulation.

As the visual algorithm of persuasion, AI images both express and distort emotions, and it is important to understand how to ethically utilize AI for civic engagement, especially following the 2024 presidential election.

I researched the emotional sentiment in AI image tweets versus non-AI image tweets. This is an important issue to address due to the spread and adoption of misinformation, the erosion of public trust, and the incitement of violence.

I started with detecting AI images and identifying superspreaders (highly influential political users) using GPT-4o and OpenCV, and for each image superspreader, I sampled 300 tweets. For each tweet, I web scraped the comments, and in total, I web scraped 10,000 political comments. Lastly, I conducted tweetNLP emotional analysis on each comment for the emotional sentiment, with the 11 emotions being anger, anticipation, disgust, fear, joy, love, optimism, pessimism, sadness, surprise, and trust.

After assessing and comparing the averages of each emotion for the AI versus non-AI images using the Mann Whitney U test, I found that the comments under AI image tweets had more positive emotions (with higher levels of joy, optimism, love, and trust) while the comments under non-AI image tweets had more negative emotions (with higher levels of anger, anticipation, disgust, fear, pessimism, sadness, and surprise). All these levels were statistically significant, with p-values less than 0.05, with the exception of surprise.