A Campaign Aide Didn’t Write That Email. A.I. Did.
The Democratic Party has begun testing the use of artificial intelligence to write first drafts of some fund-raising messages, appeals that often perform better than those written entirely by human beings.
Fake A.I. images of Donald J. Trump getting arrested in New York spread faster than they could be fact-checked last week.
And voice-cloning tools are producing vividly lifelike audio of President Biden — and many others — saying things they did not actually say.
Artificial intelligence isn’t just coming soon to the 2024 campaign trail. It’s already here.
The swift advance of A.I. promises to be as disruptive to the political sphere as to broader society. Now any amateur with a laptop can manufacture the kinds of convincing sounds and images that were once the domain of the most sophisticated digital players. This democratization of disinformation is blurring the boundaries between fact and fake at a moment when the acceptance of universal truths — that Mr. Biden beat Mr. Trump in 2020, for example — is already being strained.
And as synthetic media gets more believable, the question becomes: What happens when people can no longer trust their own eyes and ears?
Inside campaigns, artificial intelligence is expected to soon help perform mundane tasks that previously required fleets of interns. Republican and Democratic engineers alike are racing to develop tools to harness A.I. to make advertising more efficient, to engage in predictive analysis of public behavior, to write more and more personalized copy and to discover new patterns in mountains of voter data. The technology is evolving so fast that most predict a profound impact, even if specific ways in which it will upend the political system are more speculation than science.
“It’s an iPhone moment — that’s the only corollary that everybody will appreciate,” said Dan Woods, the chief technology officer on Mr. Biden’s 2020 campaign. “It’s going to take pressure testing to figure out whether it’s good or bad — and it’s probably both.”
OpenAI, whose ChatGPT chatbot ushered in the generative-text gold rush, has already released a more advanced model. Google has announced plans to expand A.I. offerings inside popular apps like Google Docs and Gmail, and is rolling out its own chatbot. Microsoft has raced a version to market, too. A smaller firm, ElevenLabs, has developed a text-to-audio tool that can mimic anyone’s voice in minutes. Midjourney, a popular A.I. art generator, can conjure hyper-realistic images with a few lines of text that are compelling enough to win art contests.
“A.I. is about to make a significant change in the 2024 election because of machine learning’s predictive ability,” said Brad Parscale, Mr. Trump’s first 2020 campaign manager, who has since founded a digital firm that advertises some A.I. capabilities.
Disinformation and “deepfakes” are the dominant fear. While forgeries are nothing new to politics — a photoshopped image of John Kerry and Jane Fonda was widely shared in 2004 — the ability to produce and share them has accelerated, with viral A.I. images of Mr. Trump being restrained by the police only the latest example. A fake image of Pope Francis in a white puffy coat went viral in recent days, as well.
A New Generation of Chatbots
A brave new world. A new crop of chatbots powered by artificial intelligence has ignited a scramble to determine whether the technology could upend the economics of the internet, turning today’s powerhouses into has-beens and creating the industry’s next giants. Here are the bots to know:
ChatGPT. ChatGPT, the artificial intelligence language model from a research lab, OpenAI, has been making headlines since November for its ability to respond to complex questions, write poetry, generate code, plan vacations and translate languages. GPT-4, the latest version introduced in mid-March, can even respond to images (and ace the Uniform Bar Exam).
Bing. Two months after ChatGPT’s debut, Microsoft, OpenAI’s primary investor and partner, added a similar chatbot, capable of having open-ended text conversations on virtually any topic, to its Bing internet search engine. But it was the bot’s occasionally inaccurate, misleading and weird responses that drew much of the attention after its release.
Bard. Google’s chatbot, called Bard, was released in March to a limited number of users in the United States and Britain. Originally conceived as a creative tool designed to draft emails and poems, it can generate ideas, write blog posts and answer questions with facts or opinions.
Ernie. The search giant Baidu unveiled China’s first major rival to ChatGPT in March. The debut of Ernie, short for Enhanced Representation through Knowledge Integration, turned out to be a flop after a promised “live” demonstration of the bot was revealed to have been recorded.
Many are particularly worried about local races, which receive far less scrutiny. Ahead of the recent primary in the Chicago mayoral race, a fake video briefly sprung up on a Twitter account called “Chicago Lakefront News” that impersonated one candidate, Paul Vallas.
“Unfortunately, I think people are going to figure out how to use this for evil faster than for improving civic life,” said Joe Rospars, who was chief strategist on Senator Elizabeth Warren’s 2020 campaign and is now the chief executive of a digital consultancy.
Those who work at the intersection of politics and technology return repeatedly to the same historical hypothetical: If the infamous “Access Hollywood” tape broke today — the one in which Mr. Trump is heard bragging about assaulting women and getting away with it — would Mr. Trump acknowledge it was him, as he did in 2016?
The nearly universal answer was no.
“I think about that example all the time,” said Matt Hodges, who was the engineering director on Mr. Biden’s 2020 campaign and is now executive director of Zinc Labs, which invests in Democratic technology. Republicans, he said, “may not use ‘fake news’ anymore. It may be ‘Woke A.I.’”
For now, the frontline function of A.I. on campaigns is expected to be writing first drafts of the unending email and text cash solicitations.
“Given the amount of rote, asinine verbiage that gets produced in politics, people will put it to work,” said Luke Thompson, a Republican political strategist.
As an experiment, The New York Times asked ChatGPT to produce a fund-raising email for Mr. Trump. The app initially said, “I cannot take political sides or promote any political agenda.” But then it immediately provided a template of a potential Trump-like email.
The chatbot denied a request to make the message “angrier” but complied when asked to “give it more edge,” to better reflect the often apocalyptic tone of Mr. Trump’s pleas. “We need your help to send a message to the radical left that we will not back down,” the revised A.I. message said. “Donate now and help us make America great again.”
Among the prominent groups that have experimented with this tool is the Democratic National Committee, according to three people briefed on the efforts. In tests, the A.I.-generated content the D.N.C. has used has, as often as not, performed as well or better than copy drafted entirely by humans, in terms of generating engagement and donations.
Party officials still make edits to the A.I. drafts, the people familiar with the efforts said, and no A.I. messages have yet been written under the name of Mr. Biden or any other person, two people said. The D.N.C. declined to comment.
Higher Ground Labs, a small venture capital firm that invests in political technology for progressives, is currently working on a project, called Quiller, to more systematically use A.I. to write, send and test the effectiveness of fund-raising emails — all at once.
“A.I. has mostly been marketing gobbledygook for the last three cycles,” said Betsy Hoover, a founding partner at Higher Ground Labs who was the director of digital organizing for President Barack Obama’s 2012 campaign. “We are at a moment now where there are things people can do that are actually helpful.”
Political operatives, several of whom were granted anonymity to discuss potentially unsavory uses of artificial intelligence they are concerned about or planning to deploy, raised a raft of possibilities.
Some feared bad actors could leverage A.I. chatbots to distract or waste a campaign’s precious staff time by pretending to be potential voters. Others floated producing deepfakes of their own candidate to generate personalized videos — thanking supporters for their donations, for example. In India, one candidate in 2020 produced a deepfake to disseminate a video of himself speaking in different languages; the technology is far superior now.
Mr. Trump himself shared an A.I. image in recent days that appeared to show him kneeling in prayer. He posted it on Truth Social, his social media site, with no explanation.
One strategist predicted that the next generation of dirty tricks could be direct-to-voter misinformation that skips social media sites entirely. What if, this strategist said, an A.I. audio recording of a candidate was sent straight to the voice mail of voters on the eve of an election?
Synthetic audio and video are already swirling online, much of it as parody.
On TikTok, there is an entire genre of videos featuring Mr. Biden, Mr. Obama and Mr. Trump profanely bantering, with the A.I.-generated audio overlaid as commentary during imaginary online video gaming sessions.
On “The Late Show,” Stephen Colbert recently used A.I. audio to have the Fox News host Tucker Carlson “read” aloud his text messages slamming Mr. Trump. Mr. Colbert labeled the audio as A.I. and the image on-screen showed a blend of Mr. Carlson’s face and a Terminator cyborg for emphasis.
The right-wing provocateur Jack Posobiec pushed out a “deepfake” video last month of Mr. Biden announcing a national draft because of the conflict in Ukraine. It was quickly seen by millions.
“The videos we’ve seen in the last few weeks are really the canary in the coal mine,” said Hany Farid, a professor of computer science at University of California at Berkeley, who specializes in digital forensics. “We measure advances now not in years but in months, and there are many months before the election.”
Some A.I. tools were deployed in 2020. The Biden campaign created a program, code-named Couch Potato, that linked facial recognition, voice-to-text and other tools to automate the transcription of live events, including debates. It replaced the work of a host of interns and aides, and was immediately searchable through an internal portal.
The technology has improved so quickly, Mr. Woods said, that off-the-shelf tools are “1,000 times better” than what had to be built from scratch four years ago.
One looming question is what campaigns can and cannot do with OpenAI’s powerful tools. One list of prohibited uses last fall lumped together “political campaigns, adult content, spam, hateful content.”
Kim Malfacini, who helped create the OpenAI’s rules and is on the company’s trust and safety team, said in an interview that “political campaigns can use our tools for campaigning purposes. But it’s the scaled use that we are trying to disallow here.” OpenAI revised its usage rules after being contacted by The Times, specifying now that “generating high volumes of campaign materials” is prohibited.
Tommy Vietor, a former spokesman for Mr. Obama, dabbled with the A.I. tool from ElevenLabs to create a faux recording of Mr. Biden calling into the popular “Pod Save America” podcast that Mr. Vietor co-hosts. He paid a few dollars and uploaded real audio of Mr. Biden, and out came an audio likeness.
“The accuracy was just uncanny,” Mr. Vietor said in an interview.
The show labeled it clearly as A.I. But Mr. Vietor could not help noticing that some online commenters nonetheless seemed confused. “I started playing with the software thinking this is so much fun, this will be a great vehicle for jokes,” he said, “and finished thinking, ‘Oh God, this is going to be a big problem.’”