Only one in four registered voters have strong confidence in their ability to tell the difference between real and AI-generated visual content, according to new research. Responses to a poll on artificial intelligence in politics also showed broad concerns among both Democrats and Republicans with AI’s influence over elections, as well as demands for increased curbs on its use. Members of both parties don’t want their candidates tweeting synthetic media.
The survey was provided exclusively to the Guardian by UK-based research firm Savanta, which originally commissioned it in response to a Guardian story about the origins of Taylor Swift deepfakes posted on Truth Social by Donald Trump. Savanta polled a representative and weighted sample of 2,004 US adults from across different demographics and regions.
“This is a very real problem, and one that voters want social media companies to grapple with,” Ethan Granholm, a research analyst at Savanta, said.
When asked about being able to tell the difference between real and AI content, 35% of Democrats responded that they were only slightly confident or not confident at all in their ability to make a correct judgment, while 45% of Republicans said they were not at all or only slightly confident in their discernment. Republicans were also more likely than Democrats – a difference of 72% to 66%, respectively – to say that it was not acceptable for political candidates to post AI-generated content without clearly labeling it.
One in four registered voters said that they were “disappointed” with the former president sharing the AI images of Swift, according to the Savanta survey. That included 23% of Republican voters who responded to the survey, highlighting what Granholm called a risk for politicians deciding to share AI-generated deepfakes.
“Former President Trump disappointed and concerned a significant proportion of his voters when he shared false imagery suggesting he had Taylor Swift’s endorsement,” Granholm said.
While the most popular option among respondents for how to deal with AI-generated content is to clearly label it, around one in four people wanted a complete ban.
AI-generated deepfakes have been a concern among researchers and election security officials for years, but the recent boom in widely accessible image generators and other AI tools has drastically lowered the bar for creating misinformation. Manipulated audio, video and images have cropped up in elections all over the world this year, with deepfakes targeting the US presidential election resulting in criminal charges and calls for more regulation.
A large majority of US voters from both parties believe that social media companies should be doing more to address AI-generated images and audio created or posted by campaigns, with 76% of people surveyed calling for more and stronger action. An even higher proportion of respondents who were above 60 years old, 83%, believed that platforms should doing more to protect and inform users.
Since the survey was conducted between 22 and 24 August, deepfakes and AI-generated content related to the election has continued to proliferate. On Twitter/X, which is owned by pro-Trump billionaire Elon Musk, AI-generated images of presidential nominees Trump and Kamala Harris have spread widely following the platform’s release of its Grok image generator, which lacks safeguards against producing deepfakes of public figures. Musk himself has shared Grok-made images of the Democratic presidential nominee as a communist.
Trump has likewise falsely accused Harris’s campaign of using AI to fake footage of a large crowd at one of her rallies. Analysis of the rally showed that Harris’s crowd was real, but the incident served as an example of politicians’ ability to dispute reality through claims that real events were the product of AI.
