Events Calendar

AI Seminar-Evaluating Text to Image Platforms Content Moderation During the 2024 US Presidential Election

Speaker: Kevin Greene, Princeton University, Princeton University

Talk Title: Evaluating Text to Image Platforms Content Moderation During the 2024 US Presidential Election

Series: AI Seminar

Abstract: Join Zoom Meeting: https://usc.zoom.us/j/96982313329?pwd=rZkvQ5qgBMsQfY3MjZgn4WrOJTOJjp.1 Meeting ID: 969 8231 3329 Passcode: 853171 Register in advance for this webinar:  https://usc.zoom.us/webinar/register/WN_erWJIqKDSzifVzFRoezM5A   How do generative AI platforms’ content moderation policies handle the creation of political deepfakes? There are considerable concerns about the risks posed by AI generated images of political leaders, but no systematic evaluation detailing how AI platforms address this outcome. We leverage an automated pipeline to extract and transform references to individuals on the US Presidential tickets from prominent media into prompts for generative AI systems, enabling politically diverse, externally valid evaluations. These prompts are sent to three prominent T2I platforms each week for the final three months of the 2024 US Presidential election. First, we show that the platforms take different approaches to content moderation. These differences contribute to there being low agreement in blocking behavior between platforms. Second, there is little consistency in the blocking behavior within platforms over time. Stability AI allowed almost all prompts featuring political figures until a sudden change two weeks before the 2024 election. Further, almost no prompts were blocked in every week of our collection. Our findings highlight the importance of developing scalable context specific approaches to monitoring text-to-image platforms.

Biography: Kevin T. Greene is an Academic Research Manager in the Empirical Studies of Conflict project at Princeton University leading the Digital Conflict and Information Integrity project. He studies the role of information and information communication technologies in international and domestic politics. Ongoing projects investigate the spread of unreliable content from algorithmic recommendations, the social and political risks posed by generative AI, and the strategies employed by foreign influence operations. His research has been funded by the National Science Foundation, Lockheed Martin Advanced Technology Labs and Microsoft and published in Science Advances, PNAS Nexus, American Political Science Review, Political Analysis, the Journal of Politics, and Political Communication among others.   His research has been funded by the National Science Foundation, Lockheed Martin Advanced Technology Labs and Microsoft and published in Science Advances, PNAS Nexus, American Political Science Review, Political Analysis, the Journal of Politics, and Political Communication among others.
 
 

Host: Zhuoyu Shi and Pete Zamar

More Info: https://www.isi.edu/events/5317/evaluating-text-to-image-platforms-content-moderation-during-the-2024-us-presidential-election/

Webcast: https://usc.zoom.us/j/96982313329?pwd=rZkvQ5qgBMsQfY3MjZgn4WrOJTOJjp.1

Location: Information Science Institute (ISI) - Virtual Only

WebCast Link: https://usc.zoom.us/j/96982313329?pwd=rZkvQ5qgBMsQfY3MjZgn4WrOJTOJjp.1

Audiences: Everyone Is Invited

Contact: Pete Zamar

Event Link: https://www.isi.edu/events/5317/evaluating-text-to-image-platforms-content-moderation-during-the-2024-us-presidential-election/