Disclosing collaboration: Visualizing human-AI collaboration in news production

This thesis explores how to disclose human-AI collaboration in news production through visual representations. As generative AI becomes more integrated into journalism, it is important to maintain trust and transparency, which are core values for news organizations. This requires transparency about both human and AI contributions. However, existing AI disclosure visualizations often fail to capture the nuanced collaboration between humans and AI, reducing complex editorial processes to overly simplistic labels. Through co-design sessions with designers (n=10), new disclosure visualizations were created to close this research gap, focusing on visualizing different ratios of human-AI collaboration. Based on prior works on visualization principles within HCI and Information Visualization, we narrowed down our set of 69 design ideas to four diverse designs that represented human-AI collaboration, and were feasible to test within the practical constraints of a controlled user study. The four designs that were turned into prototypes were a menu-based chatbot, a human-AI role-based timeline, a task-based timeline, and a standard text disclosure. These prototypes were evaluated in a lab-based user study (n=32) using behavioral measurements (incl. eye tracking) and user feedback. The study examined how visualization type and human-AI collaboration dynamics influenced perceived clarity, perceived information, perceptions of AI and journalist contributions, and visual attention patterns when engaging with news articles and disclosure visualizations. Each visualization was effective in communicating the human-AI collaboration. Specifically, while the chatbot disclosure was more suitable for providing in-depth information, the task-based and role-based timelines were better suited for providing an overview of human-AI collaboration in journalism. These findings show that the choice of representation depends on the article topic, information needs, and reader engagement. Integrating these visualizations in the journalism workflow can potentially help normalize AI use in journalism, and strengthen transparency practices by disclosing human-AI collaboration visually.