AI-Imagery Detection and Counter-Misinformation Strategies in Social Media

Central Research Questions

The proposed research focuses on understanding the potential dangers of AI-generated imagery in spreading misinformation, particularly in the context of social media and political campaigns. The research aims to address three main questions: (1) What are the cognitive mechanisms involved in detecting AI-generated images on social media? (2) Can visual watermarking mitigate the effects of the "truthiness" effect, which increases the perceived truthfulness of false information? (3) How effective are brief educational interventions in enhancing the public's ability to discern AI-generated images under the cognitive pressures typical of social media engagement?

These questions serve as the focus of the research to investigate the threat posed by AI-generated imagery in the proliferation of misinformation, while also identifying a practical solution to combat the issue. Methodology

To answer the research questions, the project is designed around three studies. The first study will involve a cross-cultural comparison between participants in the UK and Singapore, investigating differential factors that affect the ability to detect AI-generated images. Participants will view a randomized mix of AI-generated and authentic images and will be tasked with identifying which are real.

A questionnaire will follow, exploring their familiarity with AI, their cultural background, and their overall digital literacy. This study will be conducted online, leveraging platforms such as Qualtrics.

The second study will examine how cognitive load influences the ability to detect AI-generated images and assesses whether visual watermarking can disrupt the truthiness effect. Participants will be divided into groups based on cognitive load and exposed to both AI-generated and real images, some of which will feature watermarks. Thereafter, analysis will be conducted to assess whether watermarking prompts users to allocate more cognitive resources to evaluate the authenticity of images and associated claims, to reduce the risks on misinformation.

Finally, the third study will test the effectiveness of a brief educational intervention designed to help social media users detect AI-generated imagery. The intervention will consist of a short informative module highlighting key features of AI-generated images that serves to encourage users to remain vigilant when processing political information.

Participants will be divided into intervention and control groups and tasked with detecting AI-generated images under varying cognitive load conditions. This study will help assess whether such an intervention can be easily implemented during high-stakes events such as elections to safeguard against misinformation.

Details of Any Collaborations NA Envisaged Outputs

This research promises significant contributions to the field of digital media literacy and misinformation management in social media. By exploring cross-cultural differences of innate capabilities to distinguish between authentic and AI-generated images, this study will shed light on important protective or vulnerability factors to digital misinformation.

Investigation into the effects of visual watermarking will provide insight into the effectiveness and cognitive mechanisms of a common approach to ensuring authenticity. Most importantly, the assessment of a text-based advisory offers an alternate solution to the problem of large scales of misinformation. This project not only aims to advance theoretical understanding of how people interact with and interpret AI-generated imagery but also seeks to ultimately design important solutions for spread of political misinformation from social media platforms.

The findings are expected to contribute to the resilience of information ecosystems, particularly in safeguarding democratic processes and public discourse against the tide of digital misinformation.