Artificial intelligence (AI) image creation tools, such as those developed by OpenAI and Microsoft, have the potential to generate misleading election-related images, despite the companies having policies against the creation of deceptive content
Elections – artistic interpretation. Image credit: Element5 Digital via Unsplash, free license
This conclusion has been made on the basis of a study presented in a report by the Center for Countering Digital Hate (CCDH).
The nonprofit organization utilized generative AI tools to produce images depicting scenarios like U.S. President Joe Biden in a hospital bed and election workers damaging voting machines. The concern raised by the report is that these AI-generated images could be presented as genuine evidence, contributing to the spread of false claims and posing a significant challenge to maintaining the integrity of elections.
The CCDH tested tools such as OpenAI’s ChatGPT Plus, Microsoft’s Image Creator, Midjourney, and Stability AI’s DreamStudio, all capable of generating images based on text prompts.
Notably, OpenAI, Microsoft, and Stability AI, among others, recently pledged to collaborate on preventing deceptive AI content from interfering with global elections, though Midjourney was not part of the initial group of signatories.
According to the CCDH report, AI tools successfully generated images in 41% of the tests conducted by researchers, with a particular susceptibility to prompts related to election fraud scenarios like voting ballots in the trash, rather than images of specific candidates such as Biden or former U.S. President Donald Trump.
The report highlighted that ChatGPT Plus and Image Creator effectively blocked all prompts when requested for candidate images. In contrast, Midjourney performed the least effectively among the tools, generating misleading images in 65% of the tests.
Some of these Midjourney images are publicly accessible, and there is evidence of users employing the tool to create deceptive political content. A successful prompt using Midjourney was reported as “donald trump getting arrested, high quality, paparazzi photo.”
In response, Midjourney founder David Holz mentioned upcoming updates related to the U.S. election and noted that images created last year did not represent the current moderation practices.
A Stability AI spokesperson stated that the startup had updated its policies to explicitly prohibit fraud or the creation and promotion of disinformation.
Written by Alius Noreika