Popular AI image generators are reportedly giving inaccurate and erroneous results related to the 2024 U.S. Presidential Elections, according to a new study from the Center for Countering Digital Hate.
Out of the 40 text prompts given to Midjourney, ChatGPT Plus, DreamStudio and Microsoft's Image Creator, the group recorded 41% of the generated images to contain political disinformation.
(Photo : Drew Angerer/AFP via Getty Images)
The watchdog singled out Midjourney for generating the most amount of election disinformation, failing 65% of the 160 independent test runs conducted.
AI Firms Promise to Improve Image Generators Ahead of U.S. Elections
Speaking with CNN, the AI platforms said that they have already "taken steps" to improve content policies in their image generation models to prevent being used to spread political misinformation.
All four companies promised to continue updating their systems as the voting period inches closer.
The study comes after reports of AI-generated images showing former President Donald Trump, one of the presidential candidates, smiling with Black Americans went viral.
Trump's campaign team has been reportedly trying to get Black American voters to join his primarily White supporters.
Also Read: AI Chatbot 'Hallucinations' Could Affect Votes in 2024 Elections
CCDH Demands Better AI Policies Ahead of 2024 US Elections
Along with the findings, the CCDH is calling for AI companies to "provide responsible safeguards" to their technology, as well as collaborate with independent tech experts.
The group highlighted that the politically inaccurate images were generated despite anti-election disinformation policies by Midjourney, Image Creator and ChatGPT.
Development and safety tests of most generative AI remain within the company, preventing independent AI researchers from alerting them of vulnerabilities, flaws and issues before being released to the public.
With the arrival of OpenAI's video-to-text Sora AI, which can generate near-realistic videos instantly, the CCDH predicted AI-fueled disinformation would surge in the coming months if no proper protections were established.
OpenAI said it is launching a red team to moderate the content generated from Sora but has yet to deploy it.
Related Article: OpenAI's Sora Alarms Experts Over Impacts on Deepfakes, Political Propaganda
