About This Project

This research explores how visual generative AI has been utilized to spread disinformation over the past two years, identifying key topics, actors, mechanisms, and features. Additionally, it experimentally tests the impact of this content on engagement, persuasiveness, and perceived trustworthiness. Unlike previous studies that focused on potential misuse, the project analyzes real-world cases, providing practical and theoretical insights to understanding the impact of visual generative AI.

Ask the Scientists

What is the context of this research?

The rise of visual generative AI has revolutionized the visual communication landscape. These tools have facilitated the creation of high-quality, photorealistic visuals in a fraction of a second and at a minimal cost, which simplified visualizations more than ever before. This has raised huge concerns over the potential to exacerbate visual mis/disinformation. Recent data showed that over 15 billion images have been generated from text-to-image models in 2023. This figure exceeds Shutterstock’s entire image library and constitutes almost one-third of all images ever uploaded to Instagram. More recently, we have witnessed several instances where visual generative AI intensified disinformation and triggered public panic and confusion, particularly during events such as elections and ongoing wars. This underscores the urgent need for a systematic investigation into the potential harm of this content, its impact on the public, and possible solutions to face it.

What is the significance of this project?

The project provides a real-world analysis of how visual generative AI has been used to spread disinformation over the past two years. Unlike previous studies that focus on theoretical risks or hypothesized scenarios, this research systematically examines fact-checked visual generative content to identify its key topics, actors, mechanisms, and features.

Additionally, the project experimentally tests the impact of visual generative disinformation on persuasiveness, engagement, and perceived trustworthiness, offering new insights into how audiences process and react to such content. It also experimentally tests whether visual generative literacy can improve the ability to recognize and resist such content.

By providing a comprehensive, evidence-based understanding of visual generative AI and its impact, the research will inform future strategies and policies for mitigating harm, e.g., labeling generative content or training the public on identifying visual generative disinformation.

What are the goals of the project?

1. Analyze the scope of visual generative disinformation that went viral over the past two years, the actors behind it, their mechanisms, content features, main themes, and audience engagement. Visual generative disinformation flagged by fact-checking organizations will be scrapped for the analysis. Coded mechanisms will include amplification, segmentation, and obfuscation. Coded variables will include intention, target audience, frames, emotional tone, and engagement, among others. Data will be analyzed using Nvivo and SPSS.

2. Experimentally test public perceptions of this content in terms of persuasiveness, perceived trustworthiness, and potential engagement.

3. Experimentally test whether providing visual literacy training on how to identify visual generative disinformation can enhance the ability to identify and resist this content. This will help us understand the potential of such training so that the media can generalize it.