A new poll found that a majority of adults in the US think AI tools will increase the spread of false information during elections.
The warnings have grown louder and more urgent as 2024 approaches: The rapid advance of artificial intelligence (AI) tools threatens to amplify misinformation in next year’s presidential election at a scale never seen before.
Most adults in the US feel the same way, according to a new poll from The Associated Press-NORC Centre for Public Affairs Research and the University of Chicago Harris School of Public Policy.
The poll found that nearly 6 in 10 adults (58 per cent) think AI tools — which can micro-target political audiences, mass produce persuasive messages, and generate realistic fake images and videos in seconds — will increase the spread of false and misleading information during next year’s elections.
By comparison, 6 per cent think AI will decrease the spread of misinformation while one-third say it won’t make much of a difference.
“Look what happened in 2020 — and that was just social media,” said 66-year-old Rosa Rangel of Fort Worth, Texas.
Rangel, a Democrat who said she had seen a lot of “lies” on social media in 2020, said she thinks AI will make things even worse in 2024 — like a pot “brewing over.”
Just 30 per cent of American adults have used AI chatbots or image generators and fewer than half (46 per cent) have heard or read at least some about AI tools. Still, there’s a broad consensus that candidates shouldn’t be using AI.
When asked whether it would be a good or bad thing for 2024 presidential candidates to use AI in certain ways, clear majorities said it would be bad for them to create false or misleading media for political ads (83 per cent), to edit or touch-up photos or videos for political ads (66 per cent), to tailor political ads to individual voters (62 per cent) and to answer voters’ questions via chatbot (56 per cent).
The sentiments are supported by majorities of Republicans and Democrats, who agree it would be a bad thing for the presidential candidates to create false images or videos (85 per cent of Republicans and 90 per cent of Democrats) or to answer voter questions (56 per cent of Republicans and 63 per cent of Democrats).
The bipartisan pessimism toward candidates using AI comes after it already has been deployed in the Republican presidential primary.
In April, the Republican National Committee released an entirely AI-generated ad meant to show the future of the country if US President Joe Biden is reelected. It used fake but realistic-looking photos showing boarded-up storefronts, armoured military patrols in the streets and waves of immigrants creating panic. The ad disclosed in small lettering that it was generated by AI.
Ron DeSantis, the Republican governor of Florida, also used AI in his campaign for the GOP nomination. He promoted an ad that used AI-generated images to make it look as if former President Donald Trump was hugging Dr Anthony Fauci, an infectious disease specialist who oversaw the nation’s response to the COVID-19 pandemic.
Never Back Down, a super PAC supporting DeSantis, used an AI voice-cloning tool to imitate Trump’s voice, making it seem like he narrated a social media post.
“I think they should be campaigning on their merits, not their ability to strike fear into the hearts of voters,” said Andie Near, a 42-year-old from Holland, Michigan, who typically votes for Democrats.
She has used AI tools to retouch images in her work at a museum, but she said she thinks politicians using the technology to mislead can “deepen and worsen the effect that even conventional attack ads can cause.”
College student Thomas Besgen, a Republican, also disagrees with campaigns using deepfake sounds or imagery to make it seem as if a candidate said something they never said.
“Morally, that’s wrong,” the 21-year-old from Connecticut said.
Besgen, a mechanical engineering major at the University of Dayton in Ohio, said he is in favour of banning deepfake ads or, if that’s not possible, requiring them to be labelled as AI-generated.
The Federal Election Commission is currently considering a petition urging it to regulate AI-generated deepfakes in political ads ahead of the 2024 election.
While sceptical of AI’s use in politics, Besgen said he is enthusiastic about its potential for the economy and society. He is an active user of AI tools such as ChatGPT to help explain history topics he’s interested in or to brainstorm ideas. He also uses image-generators for fun — for example, to imagine what sports stadiums might look like in 100 years.
He said he typically trusts the information he gets from ChatGPT and will likely use it to learn more about the presidential candidates, something that just 5 per cent of adults say they are likely to do.
The poll found that Americans are more likely to consult the news media (46 per cent), friends and family (29 per cent), and social media (25 per cent) for information about the presidential election than AI chatbots.
“Whatever response it gives me, I would take it with a grain of salt,” Besgen said.
The vast majority of Americans are similarly sceptical toward the information AI chatbots spit out. Just 5 per cent say they are extremely or very confident that the information is factual, while 33 per cent are somewhat confident, according to the survey. Most adults (61 per cent) say they are not very or not at all confident that the information is reliable.
That’s in line with many AI experts’ warnings against using chatbots to retrieve information. The artificial intelligence large language models powering chatbots work by repeatedly selecting the most plausible next word in a sentence, which makes them good at mimicking styles of writing but also prone to making things up.
Adults associated with both major political parties are generally open to regulations on AI. They responded more positively than negatively toward various ways to ban or label AI-generated content that could be imposed by tech companies, the federal government, social media companies or the news media.
About two-thirds favour the government banning AI-generated content that contains false or misleading images from political ads, while a similar number want technology companies to label all AI-generated content made on their platforms.
Biden set in motion some federal guidelines for AI on Monday when he signed an executive order to guide the development of the rapidly progressing technology. The order requires the industry to develop safety and security standards and directs the Commerce Department to issue guidance to label and watermark AI-generated content.
Americans largely see preventing AI-generated false or misleading information during the 2024 presidential elections as a shared responsibility. About 6 in 10 (63 per cent) say a lot of the responsibility falls on the technology companies that create AI tools, but about half give a lot of that duty to the news media (53 per cent), social media companies (52 per cent), and the federal government (49 per cent).
Democrats are somewhat more likely than Republicans to say social media companies have a lot of responsibility, but generally agree on the level of responsibility for technology companies, the news media and the federal government.
The poll of 1,017 adults was conducted October 19-23, 2023, using a sample drawn from NORC’s probability-based AmeriSpeak Panel, designed to represent the US population. The margin of sampling error for all respondents is plus or minus 4.1 percentage points.