Investigating Potential Effects of Confirmation Bias on Participants’ Discernment of Real versus AI-Generated Political Images

open_access
Open Access
Author:
Chen, Claire
Area of Honors:
Psychology
Degree:
Bachelor of Science
Document Type:
Thesis
Thesis Supervisors:
Rick Gilmore, Thesis Supervisor
Rick Gilmore, Thesis Honors Advisor
Karen Gasper, Faculty Reader
Keywords:
AI
Artificial Intelligence
Confirmation Bias
Politics
Political Bias
AI Images
Fake News
AI Detection
Abstract:
Artificial intelligence (AI) has unprecedented technical abilities to create vast amounts of sophisticated media quickly and has been used in the United States to create fake images and stories. When viewing images online, Americans must now remain vigilant and discern fake images from real ones. Previous literature has discovered that participants are less discerning and more trusting when fake news headlines align with their pre-existing political beliefs, an effect of confirmation bias, but evidence remains inconclusive for studies utilizing AI-generated videos. This study examines the potential effects of confirmation bias on participants’ abilities to discern between AI-generated and real images and hypothesized that participants would more likely believe an image was real if it aligned with their political views, and political affiliation would significantly affect participants’ accuracy with images showing support for different ideologies. 126 participants were shown various (real and AI-generated) images that displayed left-leaning and right-leaning politicians in positive and negative contexts and asked to conclude if the image was real or AI-generated. Though participants were more likely to believe an image was real if it aligned politically, overall image classification accuracy remained unaffected. Older age was found to be a significant predictor of better classification accuracy, leading to new discussions about future AI education and training.