Researchers noted in a report on Wednesday that even though OpenAI and Microsoft have policies prohibiting the creation of misleading content, artificial intelligence-powered image creation tools from these companies can be used to create images that could support disinformation related to elections or voting.
Concerns about falsehoods were raised in advance of the US presidential election in November when the Centre for Countering Digital Hate (CCDH), a nonprofit that tracks online hate speech, created images of US President Joe Biden lying in a hospital bed and election workers smashing voting machines using generative AI tools.
“The potential for such AI-generated images to serve as ‘photo evidence’ could exacerbate the spread of false claims, posing a significant challenge to preserving the integrity of elections,” the study authors stated.
The ability to create images from word prompts was examined by CCDH using OpenAI’s ChatGPT Plus, Microsoft’s Image Creator, Midjourney, and Stability AI’s DreamStudio.
The claim comes after it was revealed last month that 20 tech companies, including OpenAI, Microsoft, and Stability AI, had inked a collaboration pact to stop misleading AI content from meddling in this year’s global elections. Midjourney wasn’t one of the original signatories.
According to CCDH, the AI tools produced images in 41% of the tests conducted by the researchers. They also responded best to prompts requesting images of electoral fraud, such as ballots found in the garbage, as opposed to pictures of Biden or former US President Donald Trump.
The report stated that ChatGPT Plus and Image Creator were successful in preventing any prompts when requested for candidate images.
Out of all the tools, Midjourney fared the worst, producing false photos in 65% of the tests conducted by the researchers, according to the report.
There are certain Midjourney photographs that are publicly accessible to other users, and according to CCDH, there is proof that some users have already started using the tool to produce false political content. A Midjourney user successfully utilised the prompt “donald trump getting arrested, high quality, paparazzi photo.”
“Updates related specifically to the upcoming U.S. election are coming soon,” stated Midjourney creator David Holz in an email. He also mentioned that photographs from the previous year were not indicative of the research lab’s current moderation methods.
According to a representative for Stability AI, the company revised its guidelines on Friday to forbid “fraud or the creation or promotion of disinformation.” Microsoft did not respond to a request for comment, but an OpenAI representative stated that the business was attempting to stop misuse of its products.
