May 30 (Reuters) – (Note: Paragraph 10 contains strong language)
“Actually, I really like Ron DeSantis,” Hillary Clinton revealed in a surprise online endorsement video. “He is exactly the type of player this country needs.
Joe Biden finally took off his mask and went on a cruel rant against transgender people. “You will never be a real woman,” bellows the president.
Welcome to the reality-challenged 2024 US presidential election.
Deepfakes of Clinton and Biden (realistic but fabricated videos created by AI algorithms trained on tons of online footage) have surfaced in the thousands on social media, and are threatening U.S. politics. obscuring fact and fiction in a polarized world.
Such synthetic media have been around for several years, but have been greatly enhanced over the past year by a slew of new “generative AI” tools such as Midjourney, making it cheaper and easier to create compelling deepfakes. About two people were interviewed by Reuters, with more than a dozen experts in areas such as AI, online misinformation and political activism.
“It will be very difficult for voters to tell the difference between the real and the fake,” said Darrell West, a senior researcher at Brookings University. can be imagined,” he said. Institute of Technology Innovation Center.
“Just before the election, something can fall that no one can undo.”
As the tech industry engages in an AI arms race, tools capable of generating deepfakes are released with few or incomplete guardrails to prevent harmful misinformation, the impact of technology on society. said Aza Ruskin, co-founder of the Center for Human Technology, a nonprofit that studies .
Former President Donald Trump himself, who will be competing with DeSantis and others in the Republican nomination race against Biden, posted a defaced video of CNN anchor Anderson Cooper earlier this month on his social media platform Truth. shared on social.
“That was here on CNN live on the Presidential Town Hall, and President Donald J. Trump told us a new asshole,” Cooper says in the video, but the words are his own. Not consistent with lip movements.
CNN announced that the video was a deepfake. A representative for President Trump did not respond to a request for comment on the video posted on his son Donald Jr.’s Twitter page again this week.
Major social media platforms such as Facebook, Twitter, and YouTube are working to ban and remove deepfakes, but their effectiveness in cracking down on such content has been mixed.
Deepfake Pence, not Trump
Deep Media, which develops tools to detect synthetic media, said there were three times more video deepfakes of all kinds and eight times more audio deepfakes posted online this year than the same period in 2022.
DeepMedia estimates that by 2023, about 500,000 total video and audio deepfakes will be shared on social media sites around the world. Until the end of last year, voice cloning cost $10,000 for servers and AI training, but now startups are offering it for just a few bucks, he said.
No one knows where the road for generative AI will lead, or how to effectively defend its power against mass misinformation, according to those interviewed.
Industry leader OpenAI, which has changed the game in recent months with the release of ChatGPT and updated model GPT-4, is also grappling with the problem. Chief Executive Officer Sam Altman told Congress this month that the fairness of elections was a “major concern” and called for speedy regulation of the area.
Unlike some smaller startups, OpenAI has taken steps to restrict political use of its products, according to a Reuters analysis of the terms of service of six major companies that offer generative AI services.
However, there is a gap in the guardrail.
For example, OpenAI says its image production company DALL-E is prohibited from creating public figures. In fact, when Reuters tried to create an image of Trump and Biden, the request was blocked with a message that “may not comply with our content policy.” ”
But Reuters was able to produce images of at least a dozen other US politicians, including former Vice President Mike Pence, who is also considering a 2024 White House run.
OpenAI also restricts “massive” use of its products for political purposes. This would ban the use of AI to mass send individually customized emails to voters, for example.
The Microsoft-backed company explained its political policies in an interview with Reuters, but did not respond to requests for further comment on enforcement gaps in its policies, such as blocking the image-making of politicians.
Some smaller startups have no clear limits on political content.
Midjourney, which launched last year, is a leading AI-generated image company with 16 million users on its official Discord server. The app, which ranges from free to $60 per month depending on factors such as photo volume and speed, is popular with AI designers and artists because it can generate highly realistic images of celebrities and politicians. AI researchers and AI researchers say: Interviewed by the creator.
Midjourney did not respond to a request for comment for this article. In a Discord online chat last week, CEO David Holtz said he would likely make changes before the election to combat misinformation.
Midjourney would like to collaborate on an industry solution that would allow traceability of AI-generated images using digital images equivalent to watermarks, and would also consider blocking images of political candidates, Holz said. added.
Republican AI-generated ads
While the industry struggles with how to prevent abuse, some politicians themselves are looking to harness the power of generative AI to power their campaigns.
So far, the only AI-generated political ad in the U.S. that has garnered attention is an ad published by the Republican National Committee in late April. Revealing the RNC was entirely AI-generated, this 30-second ad hints at a catastrophe scenario in which China invades Taiwan and San Francisco is shut down by crime if Biden is re-elected. I was using a fake image to do.
The RNC did not respond to a request for comment about the ad or its widespread use of AI. The Democratic National Committee declined to comment on the technology’s use.
Reuters polled all Republican presidential campaigns about their use of AI. Few responded, saying Nikki Haley’s team does not use the technology, while high-profile Perry Johnson’s camp said it uses AI to “generate and iterate copies.” but did not disclose details.
The potential for generative AI to create election campaign emails, posts and ads is attractive to some activists who feel low-cost technology could level the playing field in elections. .
Even deep in rural Hillsdale, Michigan, progress is being made in machine intelligence.
John Smith, Republican Speaker of Michigan’s 5th congressional district, has held several educational conferences to help his allies learn how to use AI for social media and ad generation.
“AI helps us play against big cats,” he said. “We see the biggest gains in local races. At 65, a farmer and county commissioner, he could easily be decapitated by a young cat with this technology.”
Political consultancies are also exploring the use of AI, further blurring the lines between reality and unreality.
Numinar Analytics, a political data company focused on Republican customers, has begun experimenting with AI content generation for audio and images, as well as voice generation that could potentially create personalized messages in the voice of a candidate, according to its founders. Will Long said in an interview.
Meanwhile, the Honan Strategy Group, a Democratic polling and strategy group, is trying to develop an AI survey bot. CEO Bradley Honan said he hopes to roll out the female bot in time for the 2023 local elections, citing research showing that both men and women are more likely to talk to female interviewers.
Reporting from San Francisco by Alexandra Ulmer and Anna Tong. Additional reporting by Christina Anagnostopoulos, Zeba Siddiqui, and Jonathan Spicer. Editing: Kenneth Li, Ross Colvin, Pravin Char
Our standards: Thomson Reuters Trust Principles.
