Political parties and campaign consultants are using artificial intelligence to build out candidate biographies and parse data with lightning-quick efficiency. Some have gone a step further. The technology is already being used to generate video advertising.
An ad posted last week on the Republican National Committee’s YouTube channel used AI technology to depict thousands of people heading toward the Golden Gate Bridge on foot due to drug-fueled crime.
But are we really on the verge of game-changing AI developments that radically alter the political horse race? The answer from most experts: maybe.
The excitement around AI and its rapid evolution is real. So are the risks. Figuring out how the technology will be implemented in the real world, however, is still in many ways a guessing game.
The Standard spoke to political organizers, data scientists and other experts about how AI is already being used in politics—and what could be coming next.
Trust but Verify
Political consultants say the most obvious progress to be found in AI comes at a basic level, such as crafting bios and rudimentary speech outlines through programs like ChatGPT. But there have also been stunning advancements in creating “deep fakes.” These sham videos, photos and audio recordings can easily avoid detection by the naked eye and undressed ear.
The deep fakes that proliferated on social media during the 2020 election were “not as scary or realistic as they are now,” said Elyse Samuels, a video reporter for The Washington Post's visual forensics team.
In a presentation at the University of Chicago’s Institute of Politics journalism conference last week, Samuels showed how the GOP’s AI-generated video of people fleeing San Francisco could be fact-checked to spot inconsistencies by cross-referencing with other verified images of the Golden Gate Bridge.
Other deep fake images that recently went viral include doctored images of Donald Trump being tackled in the streets and running from police when turning himself in on charges of falsifying business records.
“There are several different abnormalities that you can look for in an image that might be fake,” Samuels said, noting that some of the officers in the Trump images had missing or extra fingers and off-center sightlines that suggested a mash-up of multiple images.
Confirming that a deep fake is, in fact, fake takes intensive research. The Post launched its team to verify deep fakes after Russia declared war against Ukraine. A group of roughly eight reporters pored over hundreds of videos that purported to show acts of war and missile attacks, Samuels said.
Deep fakes aren’t exclusive to videos and photos, though. If AI can generate a viral new song by fake Drake and The Weeknd, the existence of the technology could also sow doubts about the veracity of real audio leaks, such as the recording of Trump making vulgar claims about groping women.
“I was talking with Hany Farid, who's a professor at UC Berkeley that does a lot of work on this, and he was saying that the group of people fighting the misinformation and trying to detect this is actually much smaller than the [number of] people making it,” Samuels said.
Ethical concerns extend beyond just deep fakes, of course. Campaign consultants are concerned that checks and balances on insensitive or racist ads could increasingly fall by the wayside in the age of AI. San Francisco Supervisor Matt Dorsey, who worked as a spokesperson and speechwriter before rising to elected office, told The Standard in an interview last year that it was his job as communications director to act as “the conscience of a campaign.”
Using AI to target voters and craft messages may be expedient, but it could also take a win-at-all-costs approach that shunts ethics to the side, said Kat Atwater, CEO of Community Tech Alliance, a data-focused group that works with progressive-leaning organizations.
“One of the things that scares me the most,” Atwater said, “especially when we’re thinking about applying AI to data science and machine learning, is the bias that may be built in, that doesn't have a check or an ethical ability to put a framework and a lens and evaluation on it.”
All About the Data
Campaign experts agree that the algorithms can be used to supercharge sophisticated data analysis at the national level, but this work is only as good as the data on which it relies.
When it comes to political campaigns, data tends to become more instructive on the state and national levels, as there are more voters to parse and more money being spent. The spending on state and federal elections in 2020 totaled $14 billion, according to Open Secrets.
Reince Priebus, the former chair of the Republican National Committee and chief of staff to President Donald Trump, is especially bullish about how much data the GOP has on voters in swing states. During a journalism conference last week at the Institute of Politics at the University of Chicago, Priebus said $120 million went toward targeting just 50,000 voters in the 2022 election for the Wisconsin U.S. Senate and governor's races.
“It's 50,000 [people], and we know, really, what beer you drink, what car you drive, how many kids you have, how much money you make, whether your mortgage is right side up or upside down,” Priebus said. “We're slicing and dicing every one of those 50,000.”
National voter files also can drill down on spending habits through data obtained from credit agencies, Priebus said, which helps develop a highly specific voter profile when targeting people in campaign calls and doorknocking. But getting this information optimized to a truly useful level with the help of AI may be a ways off.
“For years, we've had this mythology around micro-targeting. In particular, that campaigns are slicing and dicing the electorate 50 different ways and creating 50 different messages for each of those segments,” said Patrick Ruffini, a founding partner of Echelon Insights, a polling and analytics firm. “And the reality is a lot more complicated than that.”
But that isn’t to say local races here in the Bay Area haven’t been using AI.
Byron Philhour, a data scientist for San Francisco-based RTBiQ Political, said his firm, which works mostly on moderate Democrat campaigns, has been using machine learning on data sets in races as small as county supervisor campaigns.
“The problem is not with using that data to identify target groups,” Philhour said. “Really, it’s a problem with the capacity of a campaign to have the time, space, energy and money to generate very different messages for those targets.”
He added, “AI and all of this data work is a marginal improvement. It shouldn’t be, and isn’t, the central focus. People still need to make strong arguments and develop relationships.”
AI in the Short Term
Rather than replacing junior staffers, political insiders say one major benefit of AI is its ability to free people up to make personal connections with voters rather than sitting at a desk doing data entry.
Emily Norman, CEO of the Democratic Data Exchange, which compiles information for the Democratic Party’s national voter file and shares it with allied organizations, said she has two employees—one with a Ph.D. and another with a master's degree—who spend much of their time reading and classifying hundreds of thousands of answers to survey questions.
“Would I love AI to speed that up so they could be doing better things with their time?” Norman asked. “Yeah, I think that they could be doing much better things with their time.”
Another example of AI taking the leap from a helpful tool to a game-changer would be political organizing. Atwater suggested that AI could be used to “customize” the experience of likely voters from website visitors and grassroots donors to campaign workers.
“These are the kinds of things campaigns can only dream of orchestrating, that push along from interested party to donor to activist to volunteer,” Atwater said.
For now, though, the biggest benefit of AI will be confined to “the automation of grunt work,” Ruffini said.
“It's not sexy, and it's not going to be covered [by the media], but that's really what I would see as if there’s going to be an impact right in [2024],” Ruffini said. “It’ll be at the back office level, and it won't get written about unless somebody has a very clever marketing team to try to make a sexy story out of that.”